feat: 补充业务闭环与运维健康检查

ai-services/vision-inference/app/cpu_runtime.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import base64
 import json
 import os
+import sys
 import threading
 import time
 from collections import OrderedDict
@@ -17,8 +18,37 @@ import numpy as np
 
 from .schemas import Resource
 
+_DLL_DIRECTORY_HANDLES: list[Any] = []
+
+
+def _configure_windows_cuda_dll_path() -> None:
+    if os.name != "nt" or not hasattr(os, "add_dll_directory"):
+        return
+    seen: set[str] = set()
+    for entry in sys.path:
+        if not entry:
+            continue
+        nvidia_root = Path(entry) / "nvidia"
+        if not nvidia_root.is_dir():
+            continue
+        for dll_dir in nvidia_root.glob("*/bin"):
+            key = str(dll_dir.resolve())
+            if key in seen:
+                continue
+            seen.add(key)
+            try:
+                _DLL_DIRECTORY_HANDLES.append(os.add_dll_directory(key))
+                os.environ["PATH"] = key + os.pathsep + os.environ.get("PATH", "")
+            except OSError:
+                continue
+
+
+_configure_windows_cuda_dll_path()
+
 try:
     import onnxruntime as ort
+    if hasattr(ort, "preload_dlls"):
+        ort.preload_dlls()
 except ImportError:  # pragma: no cover - exercised by the lightweight source test environment
     ort = None
 
@@ -28,6 +58,44 @@ except ImportError:  # pragma: no cover - exercised by the lightweight source te
     Minio = None
 
 
+def _repo_root() -> Path | None:
+    current = Path(__file__).resolve()
+    for parent in current.parents:
+        if (parent / "infra" / "model-registry").is_dir() and (parent / "runtime").is_dir():
+            return parent
+    return None
+
+
+def _default_registry_path() -> Path:
+    configured = os.getenv("RAIL_MODEL_REGISTRY")
+    if configured:
+        return Path(configured)
+    container_path = Path("/app/config/model-registry.json")
+    if container_path.is_file():
+        return container_path
+    root = _repo_root()
+    if root:
+        local_path = root / "infra" / "model-registry" / "cpu-models.json"
+        if local_path.is_file():
+            return local_path
+    return container_path
+
+
+def _default_model_dir() -> Path:
+    configured = os.getenv("RAIL_MODEL_DIR")
+    if configured:
+        return Path(configured)
+    container_path = Path("/models")
+    if container_path.exists():
+        return container_path
+    root = _repo_root()
+    if root:
+        local_path = root / "runtime" / "models"
+        if local_path.is_dir():
+            return local_path
+    return container_path
+
+
 @dataclass(frozen=True)
 class CpuRuntimeSettings:
     profile: str
@@ -45,13 +113,13 @@ class CpuRuntimeSettings:
     def from_env(cls) -> "CpuRuntimeSettings":
         return cls(
             profile=os.getenv("RAIL_RUNTIME_PROFILE", "cpu-local"),
-            registry_path=Path(os.getenv("RAIL_MODEL_REGISTRY", "/app/config/model-registry.json")),
+            registry_path=_default_registry_path(),
-            model_dir=Path(os.getenv("RAIL_MODEL_DIR", "/models")),
+            model_dir=_default_model_dir(),
             execution_provider=os.getenv("RAIL_EXECUTION_PROVIDER", "CPUExecutionProvider"),
             intra_op_threads=max(1, int(os.getenv("RAIL_INTRA_OP_THREADS", "4"))),
             inter_op_threads=max(1, int(os.getenv("RAIL_INTER_OP_THREADS", "1"))),
             max_concurrency=max(1, int(os.getenv("RAIL_MAX_CONCURRENCY", "1"))),
-            max_loaded_models=max(1, int(os.getenv("RAIL_MAX_LOADED_MODELS", "1"))),
+            max_loaded_models=max(1, int(os.getenv("RAIL_MAX_LOADED_MODELS", "2"))),
             max_resource_bytes=max(1, int(os.getenv("RAIL_MAX_RESOURCE_MB", "64"))) * 1024 * 1024,
             fallback_mode=os.getenv("RAIL_FALLBACK_MODE", "baseline"),
         )
@@ -154,6 +222,17 @@ class ModelRegistry:
                     "parser": "paddle-detection",
                     "input_size": 640,
                 },
+                {
+                    "model_group": "vision-detector",
+                    "model_version": "traffic-yolov8n-coco",
+                    "display_name": "YOLOv8n COCO 交通演示本地 ONNX",
+                    "family": "YOLOv8",
+                    "artifact": "vision-detector/yolov8n-coco/model.onnx",
+                    "labels": "vision-detector/yolov8n-coco/labels.txt",
+                    "parser": "ultralytics-yolo",
+                    "input_size": 640,
+                    "active": False,
+                },
                 {
                     "model_group": "vision-segmenter",
                     "model_version": "cpu-v1.0.0",
@@ -166,6 +245,17 @@ class ModelRegistry:
                     "mean": [0.5, 0.5, 0.5],
                     "std": [0.5, 0.5, 0.5],
                 },
+                {
+                    "model_group": "vision-segmenter",
+                    "model_version": "traffic-yolov8n-seg-coco",
+                    "display_name": "YOLOv8n-seg COCO 交通实例分割本地 ONNX",
+                    "family": "YOLOv8-seg",
+                    "artifact": "vision-segmenter/yolov8n-seg-coco/model.onnx",
+                    "labels": "vision-segmenter/yolov8n-seg-coco/labels.txt",
+                    "parser": "ultralytics-yolo-seg",
+                    "input_size": 640,
+                    "active": False,
+                },
                 {
                     "model_group": "thermal-analyzer",
                     "model_version": "cpu-v1.0.0",
@@ -419,6 +509,10 @@ class LazyOnnxRuntime:
     ) -> list[dict[str, Any]]:
         if spec.parser == "semantic-segmentation":
             return self._parse_segmentation(outputs, spec, parameters)
+        if spec.parser == "ultralytics-yolo-seg":
+            return self._parse_ultralytics_yolo_seg(outputs, spec, transform, parameters)
+        if spec.parser == "ultralytics-yolo":
+            return self._parse_ultralytics_yolo(outputs, spec, transform, parameters)
         return self._parse_detection(outputs, spec, transform, parameters)
 
     def _parse_detection(
@@ -475,6 +569,289 @@ class LazyOnnxRuntime:
                 break
         return results
 
+    def _parse_ultralytics_yolo(
+        self,
+        outputs: list[np.ndarray],
+        spec: ModelSpec,
+        transform: dict[str, Any],
+        parameters: dict[str, Any],
+    ) -> list[dict[str, Any]]:
+        threshold = float(parameters.get("confidence_threshold", 0.35))
+        nms_threshold = float(parameters.get("nms_iou_threshold", 0.45))
+        max_detections = int(parameters.get("max_detections", 100))
+        allowed = {str(item).strip().lower() for item in parameters.get("allowed_categories", []) if str(item).strip()}
+        labels = self._labels(spec)
+        candidates = self._yolo_output_matrix(outputs)
+
+        scored: list[dict[str, Any]] = []
+        for row in candidates:
+            if row.shape[0] < 5:
+                continue
+
+            cx, cy, width, height = (float(item) for item in row[:4])
+            if labels and row.shape[0] == len(labels) + 5:
+                objectness = float(row[4])
+                class_scores = row[5:]
+                class_id = int(np.argmax(class_scores))
+                score = objectness * float(class_scores[class_id])
+            else:
+                class_scores = row[4:]
+                class_id = int(np.argmax(class_scores))
+                score = float(class_scores[class_id])
+
+            if score < threshold:
+                continue
+
+            category = labels[class_id] if 0 <= class_id < len(labels) else f"class-{class_id}"
+            if allowed and category.lower() not in allowed:
+                continue
+
+            scale = 1.0 if max(abs(cx), abs(cy), abs(width), abs(height)) <= 2.0 else None
+            if scale is None:
+                x1 = (cx - width / 2) / transform["input_w"]
+                y1 = (cy - height / 2) / transform["input_h"]
+                x2 = (cx + width / 2) / transform["input_w"]
+                y2 = (cy + height / 2) / transform["input_h"]
+            else:
+                x1 = cx - width / 2
+                y1 = cy - height / 2
+                x2 = cx + width / 2
+                y2 = cy + height / 2
+
+            bbox = [self._clip(x1), self._clip(y1), self._clip(x2), self._clip(y2)]
+            if bbox[2] <= bbox[0] or bbox[3] <= bbox[1]:
+                continue
+
+            scored.append(
+                {
+                    "category": category,
+                    "class_id": class_id,
+                    "confidence": round(float(score), 4),
+                    "bbox": bbox,
+                }
+            )
+
+        scored.sort(key=lambda item: float(item["confidence"]), reverse=True)
+        kept: list[dict[str, Any]] = []
+        for item in scored:
+            if any(item["class_id"] == kept_item["class_id"] and self._box_iou(item["bbox"], kept_item["bbox"]) > nms_threshold for kept_item in kept):
+                continue
+            kept.append(item)
+            if len(kept) >= max_detections:
+                break
+
+        return [
+            {
+                "category": str(item["category"]),
+                "confidence": float(item["confidence"]),
+                "geometry": {"type": "BBox", "coordinates": item["bbox"], "coordinate_space": "normalized"},
+                "measurements": {},
+            }
+            for item in kept
+        ]
+
+    def _parse_ultralytics_yolo_seg(
+        self,
+        outputs: list[np.ndarray],
+        spec: ModelSpec,
+        transform: dict[str, Any],
+        parameters: dict[str, Any],
+    ) -> list[dict[str, Any]]:
+        threshold = float(parameters.get("mask_threshold", parameters.get("confidence_threshold", 0.35)))
+        nms_threshold = float(parameters.get("nms_iou_threshold", 0.45))
+        max_detections = int(parameters.get("max_detections", 40))
+        mask_binary_threshold = float(parameters.get("mask_binary_threshold", 0.5))
+        minimum_area = int(parameters.get("minimum_area", 64))
+        allowed = {str(item).strip().lower() for item in parameters.get("allowed_categories", []) if str(item).strip()}
+        labels = self._labels(spec)
+        detections, prototypes = self._yolo_seg_candidates(outputs, labels, threshold, allowed, transform)
+
+        detections.sort(key=lambda item: float(item["confidence"]), reverse=True)
+        kept: list[dict[str, Any]] = []
+        for item in detections:
+            if any(item["class_id"] == kept_item["class_id"] and self._box_iou(item["bbox"], kept_item["bbox"]) > nms_threshold for kept_item in kept):
+                continue
+            kept.append(item)
+            if len(kept) >= max_detections:
+                break
+
+        if prototypes is None:
+            return []
+        if prototypes.ndim == 4:
+            prototypes = prototypes[0]
+        if prototypes.ndim != 3:
+            return []
+
+        proto_channels, proto_h, proto_w = prototypes.shape
+        results: list[dict[str, Any]] = []
+        contour_simplification = float(parameters.get("contour_simplification", 1.4))
+        for item in kept:
+            coefficients = np.asarray(item["mask_coefficients"], dtype=np.float32)
+            if coefficients.shape[0] != proto_channels:
+                continue
+            mask_logits = np.tensordot(coefficients, prototypes, axes=(0, 0))
+            mask = 1.0 / (1.0 + np.exp(-mask_logits))
+            x1, y1, x2, y2 = item["bbox"]
+            left = max(0, min(proto_w - 1, int(np.floor(x1 * proto_w))))
+            top = max(0, min(proto_h - 1, int(np.floor(y1 * proto_h))))
+            right = max(left + 1, min(proto_w, int(np.ceil(x2 * proto_w))))
+            bottom = max(top + 1, min(proto_h, int(np.ceil(y2 * proto_h))))
+            cropped = np.zeros_like(mask, dtype=np.uint8)
+            cropped[top:bottom, left:right] = (mask[top:bottom, left:right] >= mask_binary_threshold).astype(np.uint8) * 255
+            mask_area = int(np.count_nonzero(cropped))
+            if mask_area < minimum_area:
+                continue
+            contours, _ = cv2.findContours(cropped, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+            points: list[list[float]] = []
+            if contours:
+                contour = max(contours, key=cv2.contourArea)
+                epsilon = max(0.5, cv2.arcLength(contour, True) * 0.006 + contour_simplification)
+                simplified = cv2.approxPolyDP(contour, epsilon, True)
+                points = [
+                    [self._clip(float(point[0][0]) / proto_w), self._clip(float(point[0][1]) / proto_h)]
+                    for point in simplified
+                ]
+                if len(points) >= 3:
+                    points.append(points[0])
+                else:
+                    points = []
+            results.append(
+                {
+                    "category": str(item["category"]),
+                    "confidence": round(float(item["confidence"]), 4),
+                    "geometry": {"type": "Polygon", "coordinates": [points], "coordinate_space": "normalized"},
+                    "measurements": {"area_ratio": round(mask_area / float(proto_h * proto_w), 6)},
+                    "mask": self._encode_binary_mask(cropped),
+                }
+            )
+        return results
+
+    def _yolo_seg_candidates(
+        self,
+        outputs: list[np.ndarray],
+        labels: list[str],
+        threshold: float,
+        allowed: set[str],
+        transform: dict[str, Any],
+    ) -> tuple[list[dict[str, Any]], np.ndarray | None]:
+        feature_groups: dict[tuple[int, int], dict[str, np.ndarray]] = {}
+        prototypes: np.ndarray | None = None
+        class_count = len(labels) or 80
+        mask_channels = 32
+
+        for output in outputs:
+            array = np.asarray(output)
+            if array.ndim == 4 and array.shape[0] == 1:
+                array = array[0]
+            if array.ndim != 3:
+                continue
+            if array.shape[0] == mask_channels and array.shape[1] >= 80 and array.shape[2] >= 80:
+                prototypes = array.astype(np.float32, copy=False)
+                continue
+            if array.shape[-1] in {64, class_count, mask_channels}:
+                height, width, channels = array.shape
+                entry = feature_groups.setdefault((height, width), {})
+                if channels == 64:
+                    entry["boxes"] = array.astype(np.float32, copy=False)
+                elif channels == class_count:
+                    entry["classes"] = array.astype(np.float32, copy=False)
+                elif channels == mask_channels:
+                    entry["masks"] = array.astype(np.float32, copy=False)
+
+        candidates: list[dict[str, Any]] = []
+        for (height, width), group in feature_groups.items():
+            if not {"boxes", "classes", "masks"}.issubset(group):
+                continue
+            stride_x = transform["input_w"] / width
+            stride_y = transform["input_h"] / height
+            boxes = self._decode_yolo_dfl(group["boxes"], stride_x, stride_y)
+            class_scores = self._sigmoid(group["classes"].reshape((-1, class_count)))
+            mask_coefficients = group["masks"].reshape((-1, mask_channels))
+            best_ids = np.argmax(class_scores, axis=1)
+            best_scores = class_scores[np.arange(class_scores.shape[0]), best_ids]
+            selected_indices = np.where(best_scores >= threshold)[0]
+
+            for index in selected_indices:
+                class_id = int(best_ids[index])
+                category = labels[class_id] if 0 <= class_id < len(labels) else f"class-{class_id}"
+                if allowed and category.lower() not in allowed:
+                    continue
+                x1, y1, x2, y2 = boxes[index]
+                bbox = [
+                    self._clip(float(x1) / transform["input_w"]),
+                    self._clip(float(y1) / transform["input_h"]),
+                    self._clip(float(x2) / transform["input_w"]),
+                    self._clip(float(y2) / transform["input_h"]),
+                ]
+                if bbox[2] <= bbox[0] or bbox[3] <= bbox[1]:
+                    continue
+                candidates.append(
+                    {
+                        "category": category,
+                        "class_id": class_id,
+                        "confidence": float(best_scores[index]),
+                        "bbox": bbox,
+                        "mask_coefficients": mask_coefficients[index],
+                    }
+                )
+        return candidates, prototypes
+
+    def _decode_yolo_dfl(self, boxes: np.ndarray, stride_x: float, stride_y: float) -> np.ndarray:
+        height, width, _ = boxes.shape
+        reg_max = boxes.shape[-1] // 4
+        distribution = boxes.reshape((height, width, 4, reg_max))
+        distribution = self._softmax(distribution, axis=-1)
+        projection = np.arange(reg_max, dtype=np.float32)
+        distances = np.sum(distribution * projection, axis=-1)
+        grid_y, grid_x = np.meshgrid(np.arange(height, dtype=np.float32), np.arange(width, dtype=np.float32), indexing="ij")
+        center_x = (grid_x + 0.5) * stride_x
+        center_y = (grid_y + 0.5) * stride_y
+        left = distances[..., 0] * stride_x
+        top = distances[..., 1] * stride_y
+        right = distances[..., 2] * stride_x
+        bottom = distances[..., 3] * stride_y
+        decoded = np.stack([center_x - left, center_y - top, center_x + right, center_y + bottom], axis=-1)
+        return decoded.reshape((-1, 4))
+
+    def _yolo_output_matrix(self, outputs: list[np.ndarray]) -> np.ndarray:
+        best: np.ndarray | None = None
+        for output in outputs:
+            array = np.squeeze(np.asarray(output))
+            if array.ndim != 2:
+                continue
+            if array.shape[0] < array.shape[1] and array.shape[0] <= 512:
+                array = array.T
+            if array.shape[-1] < 5:
+                continue
+            if best is None or array.shape[0] > best.shape[0]:
+                best = array.astype(np.float32, copy=False)
+        if best is None:
+            raise ValueError("模型输出中没有可识别的 YOLO 检测矩阵")
+        return best
+
+    @staticmethod
+    def _box_iou(left: list[float], right: list[float]) -> float:
+        x1 = max(left[0], right[0])
+        y1 = max(left[1], right[1])
+        x2 = min(left[2], right[2])
+        y2 = min(left[3], right[3])
+        intersection = max(0.0, x2 - x1) * max(0.0, y2 - y1)
+        if intersection <= 0:
+            return 0.0
+        left_area = max(0.0, left[2] - left[0]) * max(0.0, left[3] - left[1])
+        right_area = max(0.0, right[2] - right[0]) * max(0.0, right[3] - right[1])
+        return intersection / max(1e-9, left_area + right_area - intersection)
+
+    @staticmethod
+    def _sigmoid(array: np.ndarray) -> np.ndarray:
+        return 1.0 / (1.0 + np.exp(-array))
+
+    @staticmethod
+    def _softmax(array: np.ndarray, axis: int) -> np.ndarray:
+        shifted = array - np.max(array, axis=axis, keepdims=True)
+        exp = np.exp(shifted)
+        return exp / np.sum(exp, axis=axis, keepdims=True)
+
     def _parse_segmentation(
         self, outputs: list[np.ndarray], spec: ModelSpec, parameters: dict[str, Any]
     ) -> list[dict[str, Any]]:
@@ -495,20 +872,26 @@ class LazyOnnxRuntime:
         results: list[dict[str, Any]] = []
         for class_id in (int(item) for item in np.unique(mask) if int(item) != 0):
             binary = np.where(mask == class_id, 255, 0).astype(np.uint8)
-            contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+            component_count, component_labels, stats, _ = cv2.connectedComponentsWithStats(binary, 8)
-            for contour in sorted(contours, key=cv2.contourArea, reverse=True):
+            for component_id in range(1, component_count):
-                area = cv2.contourArea(contour)
+                area = int(stats[component_id, cv2.CC_STAT_AREA])
                 if area < minimum_area:
                     continue
+                component_mask = np.where(component_labels == component_id, 255, 0).astype(np.uint8)
+                contours, _ = cv2.findContours(component_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+                if not contours:
+                    continue
+                contour = max(contours, key=cv2.contourArea)
                 epsilon = float(parameters.get("contour_simplification", 0.8))
                 simplified = cv2.approxPolyDP(contour, max(0.5, epsilon), True)
                 points = [
                     [self._clip(float(point[0][0]) / mask.shape[1]), self._clip(float(point[0][1]) / mask.shape[0])]
                     for point in simplified
                 ]
-                if len(points) < 3:
+                if len(points) >= 3:
-                    continue
+                    points.append(points[0])
-                points.append(points[0])
+                else:
+                    points = []
                 category = labels[class_id] if class_id < len(labels) else f"class-{class_id}"
                 results.append(
                     {
@@ -516,10 +899,34 @@ class LazyOnnxRuntime:
                         "confidence": 1.0,
                         "geometry": {"type": "Polygon", "coordinates": [points], "coordinate_space": "normalized"},
                         "measurements": {"area_ratio": round(area / mask.size, 6)},
+                        "mask": self._encode_binary_mask(component_mask),
                     }
                 )
         return results
 
+    @staticmethod
+    def _encode_binary_mask(mask: np.ndarray) -> dict[str, Any]:
+        binary = (mask > 0).astype(np.uint8, copy=False)
+        flat = binary.reshape(-1)
+        counts: list[int] = []
+        current = 0
+        run_length = 0
+        for value in flat:
+            item = int(value)
+            if item == current:
+                run_length += 1
+                continue
+            counts.append(run_length)
+            run_length = 1
+            current = item
+        counts.append(run_length)
+        return {
+            "encoding": "rle",
+            "width": int(binary.shape[1]),
+            "height": int(binary.shape[0]),
+            "counts": counts,
+        }
+
     def _labels(self, spec: ModelSpec) -> list[str]:
         path = spec.labels_path(self.settings.model_dir)
         if path and path.is_file():
@@ -581,6 +988,19 @@ class CpuVisionRuntime:
             )
         return self.models.predict(model_group, image, parameters, model_version)
 
+    def infer_image_array(
+        self,
+        image: np.ndarray,
+        model_group: str,
+        parameters: dict[str, Any],
+        model_version: str | None = None,
+    ) -> RuntimeOutcome:
+        if image.ndim == 2:
+            image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
+        if image.ndim == 3 and image.shape[2] == 4:
+            image = cv2.cvtColor(image, cv2.COLOR_BGRA2BGR)
+        return self.models.predict(model_group, image, parameters, model_version)
+
     def status(self) -> dict[str, Any]:
         return self.models.status()
 

ai-services/vision-inference/app/frame_runtime.py

@@ -0,0 +1,337 @@
+from __future__ import annotations
+
+import time
+from typing import Any
+
+import cv2
+import numpy as np
+
+from .cpu_runtime import CpuVisionRuntime, RuntimeOutcome
+
+
+TRAFFIC_ALLOWED_CATEGORIES = [
+    "person",
+    "bicycle",
+    "car",
+    "motorcycle",
+    "bus",
+    "train",
+    "truck",
+    "traffic light",
+    "stop sign",
+]
+
+DETECTION_SCENES: dict[str, dict[str, Any]] = {
+    "inspection": {
+        "id": "inspection",
+        "label": "铁路/无人机巡检",
+        "description": "使用项目默认可见光检测模型，适合铁路巡检、小目标异物等场景。",
+        "detection_model_version": None,
+        "segmentation_model_version": None,
+        "allowed_categories": [],
+        "default_confidence_threshold": 0.45,
+    },
+    "traffic-driving": {
+        "id": "traffic-driving",
+        "label": "驾车/道路交通",
+        "description": "使用 YOLOv8n/YOLOv8n-seg COCO 交通演示模型，过滤车辆、行人、信号灯等类别。",
+        "detection_model_version": "traffic-yolov8n-coco",
+        "segmentation_model_version": "traffic-yolov8n-seg-coco",
+        "allowed_categories": TRAFFIC_ALLOWED_CATEGORIES,
+        "default_confidence_threshold": 0.35,
+    },
+}
+
+
+class FrameInferenceService:
+    def __init__(self, runtime: CpuVisionRuntime):
+        self.runtime = runtime
+
+    @staticmethod
+    def detection_scenes() -> list[dict[str, Any]]:
+        return list(DETECTION_SCENES.values())
+
+    def decode_image(self, payload: bytes) -> np.ndarray:
+        image = cv2.imdecode(np.frombuffer(payload, dtype=np.uint8), cv2.IMREAD_UNCHANGED)
+        if image is None:
+            raise ValueError("上传帧不是可解码图像")
+        if image.ndim == 2:
+            image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
+        if image.ndim == 3 and image.shape[2] == 4:
+            image = cv2.cvtColor(image, cv2.COLOR_BGRA2BGR)
+        return image
+
+    def resize_for_inference(self, image: np.ndarray, max_width: int | None) -> np.ndarray:
+        if not max_width or max_width <= 0 or image.shape[1] <= max_width:
+            return image
+        scale = max_width / image.shape[1]
+        target = (max_width, max(1, int(round(image.shape[0] * scale))))
+        return cv2.resize(image, target, interpolation=cv2.INTER_AREA)
+
+    def infer_frame(
+        self,
+        image: np.ndarray,
+        *,
+        detect_enabled: bool,
+        segment_enabled: bool,
+        confidence_threshold: float,
+        mask_threshold: float,
+        max_detections: int,
+        max_inference_width: int | None = None,
+        detection_scene: str | None = None,
+        detection_model_version: str | None = None,
+        segmentation_model_version: str | None = None,
+    ) -> dict[str, Any]:
+        started = time.perf_counter()
+        inference_image = self.resize_for_inference(image, max_inference_width)
+        detections: list[dict[str, Any]] = []
+        segments: list[dict[str, Any]] = []
+        warnings: list[dict[str, Any]] = []
+        runtimes: dict[str, Any] = {}
+        scene = self._detection_scene(detection_scene)
+        selected_detection_version = detection_model_version or scene.get("detection_model_version")
+        selected_segmentation_version = segmentation_model_version or scene.get("segmentation_model_version")
+
+        if detect_enabled:
+            detection_parameters: dict[str, Any] = {
+                "confidence_threshold": confidence_threshold,
+                "max_detections": max_detections,
+            }
+            if scene.get("allowed_categories"):
+                detection_parameters["allowed_categories"] = scene["allowed_categories"]
+            outcome = self.runtime.infer_image_array(
+                inference_image,
+                "vision-detector",
+                detection_parameters,
+                selected_detection_version,
+            )
+            detections = self._detections(outcome)
+            runtimes["detection_latency_ms"] = outcome.latency_ms
+            if outcome.reason and not detections:
+                detections = self._demo_detections(inference_image, confidence_threshold, max_detections)
+                warnings.extend(self._fallback_warning("vision-detector", outcome.reason))
+            else:
+                warnings.extend(self._warnings("vision-detector", outcome))
+
+        if segment_enabled and scene["id"] == "traffic-driving" and not selected_segmentation_version:
+            warnings.append(
+                {
+                    "code": "TRAFFIC_SEGMENTER_NOT_INSTALLED",
+                    "message": "驾车场景当前仅安装交通目标检测 ONNX；未安装道路/车道线分割 ONNX，已跳过分割以避免误标。",
+                    "model_group": "vision-segmenter",
+                }
+            )
+        elif segment_enabled:
+            outcome = self.runtime.infer_image_array(
+                inference_image,
+                "vision-segmenter",
+                {"mask_threshold": mask_threshold, "minimum_area": 64, "contour_simplification": 0.8},
+                selected_segmentation_version,
+            )
+            segments = self._segments(outcome)
+            runtimes["segmentation_latency_ms"] = outcome.latency_ms
+            if outcome.reason and not segments:
+                segments = self._demo_segments(inference_image, mask_threshold)
+                warnings.extend(self._fallback_warning("vision-segmenter", outcome.reason))
+            else:
+                warnings.extend(self._warnings("vision-segmenter", outcome))
+
+        return {
+            "source": {"width": int(image.shape[1]), "height": int(image.shape[0])},
+            "inference": {"width": int(inference_image.shape[1]), "height": int(inference_image.shape[0])},
+            "runtime": {
+                **runtimes,
+                "total_latency_ms": round((time.perf_counter() - started) * 1000, 2),
+                "provider": self.runtime.settings.execution_provider,
+            },
+            "scene": {
+                "id": scene["id"],
+                "label": scene["label"],
+                "detection_model_version": selected_detection_version,
+                "segmentation_model_version": selected_segmentation_version,
+            },
+            "results": {"detections": detections, "segments": segments},
+            "warnings": warnings,
+        }
+
+    def _detection_scene(self, detection_scene: str | None) -> dict[str, Any]:
+        if detection_scene and detection_scene in DETECTION_SCENES:
+            return DETECTION_SCENES[detection_scene]
+        return DETECTION_SCENES["inspection"]
+
+    def _detections(self, outcome: RuntimeOutcome) -> list[dict[str, Any]]:
+        results: list[dict[str, Any]] = []
+        for item in outcome.results:
+            geometry = dict(item.get("geometry", {}))
+            if geometry.get("type") != "BBox":
+                continue
+            results.append(
+                {
+                    "category": str(item.get("category", "target")),
+                    "confidence": float(item.get("confidence", 0)),
+                    "bbox": list(geometry.get("coordinates", [])),
+                    "model_group": outcome.model.get("model_group", "vision-detector"),
+                    "model_version": outcome.model.get("model_version"),
+                    "execution_mode": outcome.execution_mode,
+                }
+            )
+        return results
+
+    def _segments(self, outcome: RuntimeOutcome) -> list[dict[str, Any]]:
+        results: list[dict[str, Any]] = []
+        for item in outcome.results:
+            geometry = dict(item.get("geometry", {}))
+            if geometry.get("type") != "Polygon":
+                continue
+            coordinates = geometry.get("coordinates") or []
+            polygon = coordinates[0] if coordinates else []
+            segment = {
+                "category": str(item.get("category", "segment")),
+                "confidence": float(item.get("confidence", 1.0)),
+                "polygon": polygon,
+                "area_ratio": dict(item.get("measurements", {})).get("area_ratio"),
+                "model_group": outcome.model.get("model_group", "vision-segmenter"),
+                "model_version": outcome.model.get("model_version"),
+                "execution_mode": outcome.execution_mode,
+            }
+            mask = item.get("mask")
+            if isinstance(mask, dict):
+                segment["mask"] = mask
+            results.append(segment)
+        return results
+
+    def _warnings(self, model_group: str, outcome: RuntimeOutcome) -> list[dict[str, Any]]:
+        if outcome.reason:
+            code = "MODEL_ARTIFACT_MISSING" if "未安装" in outcome.reason else "MODEL_INFERENCE_UNAVAILABLE"
+            return [{"code": code, "message": outcome.reason, "model_group": model_group}]
+        return []
+
+    def _fallback_warning(self, model_group: str, reason: str) -> list[dict[str, Any]]:
+        label = "目标检测" if model_group == "vision-detector" else "图像分割"
+        return [
+            {
+                "code": "DEMO_FALLBACK_ACTIVE",
+                "message": f"真实 {label} ONNX 模型未安装，已启用 OpenCV 本地演示模式；安装 model.onnx 后会自动切换真实模型。",
+                "model_group": model_group,
+                "detail": reason,
+            }
+        ]
+
+    def _demo_detections(self, image: np.ndarray, threshold: float, max_detections: int) -> list[dict[str, Any]]:
+        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+        blurred = cv2.GaussianBlur(gray, (5, 5), 0)
+        edges = cv2.Canny(blurred, 50, 140)
+        kernel = np.ones((5, 5), dtype=np.uint8)
+        mask = cv2.dilate(edges, kernel, iterations=2)
+        contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+        height, width = image.shape[:2]
+        image_area = float(width * height)
+        results: list[dict[str, Any]] = []
+
+        for contour in sorted(contours, key=cv2.contourArea, reverse=True):
+            area = cv2.contourArea(contour)
+            if area < image_area * 0.004 or area > image_area * 0.55:
+                continue
+            x, y, box_w, box_h = cv2.boundingRect(contour)
+            if box_w < 12 or box_h < 12:
+                continue
+            extent = min(1.0, area / max(1.0, box_w * box_h))
+            score = round(max(threshold, min(0.92, 0.48 + extent * 0.32 + min(0.12, area / image_area))), 4)
+            results.append(
+                {
+                    "category": "visual-target",
+                    "confidence": score,
+                    "bbox": [round(x / width, 6), round(y / height, 6), round((x + box_w) / width, 6), round((y + box_h) / height, 6)],
+                    "model_group": "opencv-demo-detector",
+                    "model_version": "local-demo",
+                    "execution_mode": "opencv-demo",
+                }
+            )
+            if len(results) >= max_detections:
+                break
+
+        if results:
+            return results
+
+        # Keep the demo visibly responsive on very smooth frames without pretending this is a trained detector.
+        return [
+            {
+                "category": "frame-region",
+                "confidence": round(max(threshold, 0.5), 4),
+                "bbox": [0.32, 0.28, 0.68, 0.72],
+                "model_group": "opencv-demo-detector",
+                "model_version": "local-demo",
+                "execution_mode": "opencv-demo",
+            }
+        ]
+
+    def _demo_segments(self, image: np.ndarray, mask_threshold: float) -> list[dict[str, Any]]:
+        height, width = image.shape[:2]
+        max_side = 360
+        scale = min(1.0, max_side / max(width, height))
+        sample = cv2.resize(image, (max(1, int(width * scale)), max(1, int(height * scale))), interpolation=cv2.INTER_AREA)
+        lab = cv2.cvtColor(sample, cv2.COLOR_BGR2LAB)
+        pixels = lab.reshape((-1, 3)).astype(np.float32)
+        criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 12, 1.0)
+        _, labels, centers = cv2.kmeans(pixels, 4, None, criteria, 2, cv2.KMEANS_PP_CENTERS)
+        label_image = labels.reshape(sample.shape[:2])
+
+        results: list[dict[str, Any]] = []
+        sample_area = float(sample.shape[0] * sample.shape[1])
+        ranked_ids = sorted(range(len(centers)), key=lambda idx: float(centers[idx][1] + centers[idx][2]), reverse=True)
+        for cluster_id in ranked_ids:
+            binary = np.where(label_image == cluster_id, 255, 0).astype(np.uint8)
+            binary = cv2.morphologyEx(binary, cv2.MORPH_OPEN, np.ones((3, 3), dtype=np.uint8))
+            binary = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, np.ones((7, 7), dtype=np.uint8))
+            contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+            for contour in sorted(contours, key=cv2.contourArea, reverse=True):
+                area = cv2.contourArea(contour)
+                if area < sample_area * 0.015 or area > sample_area * 0.75:
+                    continue
+                component_mask = np.zeros_like(binary)
+                cv2.drawContours(component_mask, [contour], -1, 255, -1)
+                epsilon = max(2.0, cv2.arcLength(contour, True) * 0.012)
+                simplified = cv2.approxPolyDP(contour, epsilon, True)
+                if len(simplified) < 3:
+                    continue
+                polygon = [
+                    [round(float(point[0][0]) / sample.shape[1], 6), round(float(point[0][1]) / sample.shape[0], 6)]
+                    for point in simplified
+                ]
+                if len(polygon) >= 3:
+                    polygon.append(polygon[0])
+                else:
+                    polygon = []
+                results.append(
+                    {
+                        "category": "visual-region",
+                        "confidence": round(max(mask_threshold, 0.62), 4),
+                        "polygon": polygon,
+                        "area_ratio": round(area / sample_area, 6),
+                        "mask": self._encode_binary_mask(component_mask),
+                        "model_group": "opencv-demo-segmenter",
+                        "model_version": "local-demo",
+                        "execution_mode": "opencv-demo",
+                    }
+                )
+                if len(results) >= 4:
+                    return results
+        return results
+
+    @staticmethod
+    def _encode_binary_mask(mask: np.ndarray) -> dict[str, Any]:
+        binary = (mask > 0).astype(np.uint8, copy=False)
+        flat = binary.reshape(-1)
+        counts: list[int] = []
+        current = 0
+        run_length = 0
+        for value in flat:
+            item = int(value)
+            if item == current:
+                run_length += 1
+                continue
+            counts.append(run_length)
+            run_length = 1
+            current = item
+        counts.append(run_length)
+        return {"encoding": "rle", "width": int(binary.shape[1]), "height": int(binary.shape[0]), "counts": counts}

ai-services/vision-inference/app/main.py

@@ -10,9 +10,14 @@ from .schemas import (
     ModelTestInferenceRequest,
     ModelTestInferenceResponse,
 )
+from .video_demo_routes import create_video_demo_router
+from .video_export_runtime import VideoExportManager
 
 app = FastAPI(title="Rail UAV Vision Inference Service", version="0.1.0")
 engine = VisionInferenceEngine()
+video_export_manager = VideoExportManager(engine.runtime)
+app.state.video_export_manager = video_export_manager
+app.include_router(create_video_demo_router(engine, video_export_manager))
 
 
 @app.get("/health", response_model=HealthResponse)

ai-services/vision-inference/app/video_demo_routes.py

@@ -0,0 +1,211 @@
+from __future__ import annotations
+
+import subprocess
+from typing import Annotated
+
+from fastapi import APIRouter, BackgroundTasks, File, Form, HTTPException, UploadFile
+from fastapi.responses import FileResponse
+
+from .engines import VisionInferenceEngine
+from .frame_runtime import FrameInferenceService
+from .video_demo_schemas import (
+    VideoExportJobCreateResponse,
+    VideoExportJobStatus,
+    WarmupRequest,
+)
+from .video_export_runtime import (
+    VIDEO_MIME_TYPES_BY_SUFFIX,
+    VideoExportManager,
+    VideoExportStateError,
+    VideoUploadError,
+    max_video_upload_bytes,
+)
+
+
+MAX_FRAME_UPLOAD_BYTES = 2 * 1024 * 1024
+
+
+def create_video_demo_router(
+    engine: VisionInferenceEngine,
+    export_manager: VideoExportManager | None = None,
+) -> APIRouter:
+    router = APIRouter(prefix="/api/v1/video-demo", tags=["video-demo"])
+    frame_service = FrameInferenceService(engine.runtime)
+    export_manager = export_manager or VideoExportManager(engine.runtime)
+
+    @router.get("/capabilities")
+    def capabilities() -> dict:
+        runtime = engine.runtime_status()
+        accelerated = str(runtime.get("execution_provider", "")).lower() in {"cudaexecutionprovider", "tensorrtexecutionprovider"}
+        return {
+            "runtime": runtime.get("runtime"),
+            "execution_provider": runtime.get("execution_provider"),
+            "execution_provider_ready": runtime.get("execution_provider_ready"),
+            "accelerated": accelerated and bool(runtime.get("execution_provider_ready")),
+            "runtime_available": runtime.get("runtime_available"),
+            "available_providers": runtime.get("available_providers", []),
+            "gpu": _gpu_info(),
+            "models": runtime.get("models", []),
+            "detection_scenes": frame_service.detection_scenes(),
+            "recommended": {
+                "max_inference_width": 960 if accelerated else 640,
+                "detection_fps": 8 if accelerated else 2,
+                "segmentation_fps": 3 if accelerated else 1,
+            },
+            "export": {
+                "enabled": True,
+                "output_root": str(export_manager.output_root),
+                "max_video_upload_bytes": max_video_upload_bytes(),
+                "allowed_extensions": sorted(VIDEO_MIME_TYPES_BY_SUFFIX),
+            },
+        }
+
+    @router.post("/warmup")
+    def warmup(request: WarmupRequest) -> dict:
+        loaded = []
+        warnings = []
+        for model_group in request.models:
+            try:
+                model_version = request.model_versions.get(model_group, request.model_version)
+                loaded.append(engine.load(model_group, model_version))
+            except Exception as exc:
+                warnings.append({"code": "MODEL_WARMUP_FAILED", "model_group": model_group, "message": str(exc)})
+        return {"loaded": loaded, "warnings": warnings, "runtime": engine.runtime_status()}
+
+    @router.post("/infer-frame")
+    async def infer_frame(
+        frame: Annotated[UploadFile, File()],
+        session_id: Annotated[str, Form()] = "default",
+        timestamp_ms: Annotated[float, Form()] = 0,
+        source_width: Annotated[int | None, Form()] = None,
+        source_height: Annotated[int | None, Form()] = None,
+        detect_enabled: Annotated[bool, Form()] = True,
+        segment_enabled: Annotated[bool, Form()] = False,
+        confidence_threshold: Annotated[float, Form()] = 0.45,
+        mask_threshold: Annotated[float, Form()] = 0.5,
+        max_detections: Annotated[int, Form()] = 100,
+        max_inference_width: Annotated[int, Form()] = 960,
+        detection_scene: Annotated[str, Form()] = "inspection",
+        detection_model_version: Annotated[str | None, Form()] = None,
+        segmentation_model_version: Annotated[str | None, Form()] = None,
+    ) -> dict:
+        payload = await frame.read(MAX_FRAME_UPLOAD_BYTES + 1)
+        max_bytes = MAX_FRAME_UPLOAD_BYTES
+        if len(payload) > max_bytes:
+            raise HTTPException(status_code=413, detail=f"单帧大小超过 {max_bytes // (1024 * 1024)}MB")
+        try:
+            image = frame_service.decode_image(payload)
+            response = frame_service.infer_frame(
+                image,
+                detect_enabled=detect_enabled,
+                segment_enabled=segment_enabled,
+                confidence_threshold=confidence_threshold,
+                mask_threshold=mask_threshold,
+                max_detections=max_detections,
+                max_inference_width=max_inference_width,
+                detection_scene=detection_scene,
+                detection_model_version=detection_model_version,
+                segmentation_model_version=segmentation_model_version,
+            )
+        except ValueError as exc:
+            raise HTTPException(status_code=400, detail=str(exc)) from exc
+        response.update(
+            {
+                "session_id": session_id,
+                "timestamp_ms": timestamp_ms,
+                "frame_id": f"{session_id}:{round(timestamp_ms)}",
+            }
+        )
+        if source_width and source_height:
+            response["source"]["reported_width"] = source_width
+            response["source"]["reported_height"] = source_height
+        return response
+
+    @router.post("/export-jobs", response_model=VideoExportJobCreateResponse)
+    async def create_export_job(
+        background_tasks: BackgroundTasks,
+        video: Annotated[UploadFile, File()],
+        detect_enabled: Annotated[bool, Form()] = True,
+        segment_enabled: Annotated[bool, Form()] = False,
+        confidence_threshold: Annotated[float, Form()] = 0.45,
+        mask_threshold: Annotated[float, Form()] = 0.5,
+        max_inference_width: Annotated[int, Form()] = 960,
+        output_fps_policy: Annotated[str, Form()] = "source",
+        analysis_stride: Annotated[int, Form()] = 1,
+        reuse_last_result: Annotated[bool, Form()] = True,
+        max_detections: Annotated[int, Form()] = 100,
+        detection_scene: Annotated[str, Form()] = "inspection",
+        detection_model_version: Annotated[str | None, Form()] = None,
+        segmentation_model_version: Annotated[str | None, Form()] = None,
+    ) -> dict:
+        if output_fps_policy not in {"source", "fixed"}:
+            raise HTTPException(status_code=400, detail="output_fps_policy 仅支持 source 或 fixed")
+        if not detect_enabled and not segment_enabled:
+            raise HTTPException(status_code=400, detail="至少需要开启目标检测或图像分割")
+        try:
+            job = await export_manager.create_job(
+                video,
+                detect_enabled=detect_enabled,
+                segment_enabled=segment_enabled,
+                confidence_threshold=confidence_threshold,
+                mask_threshold=mask_threshold,
+                max_inference_width=max_inference_width,
+                analysis_stride=analysis_stride,
+                reuse_last_result=reuse_last_result,
+                max_detections=max_detections,
+                detection_scene=detection_scene,
+                detection_model_version=detection_model_version,
+                segmentation_model_version=segmentation_model_version,
+            )
+        except VideoUploadError as exc:
+            raise HTTPException(status_code=exc.status_code, detail=str(exc)) from exc
+        background_tasks.add_task(export_manager.run_job, job["run_id"])
+        return {
+            **job,
+            "status_url": f"/api/v1/video-demo/export-jobs/{job['run_id']}",
+        }
+
+    @router.get("/export-jobs/{run_id}", response_model=VideoExportJobStatus)
+    def export_job_status(run_id: str) -> dict:
+        state = export_manager.status(run_id)
+        if state is None:
+            raise HTTPException(status_code=404, detail="导出任务不存在")
+        return state
+
+    @router.post("/export-jobs/{run_id}/cancel", response_model=VideoExportJobStatus)
+    def cancel_export_job(run_id: str) -> dict:
+        try:
+            state = export_manager.cancel(run_id)
+        except VideoExportStateError as exc:
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
+        if state is None:
+            raise HTTPException(status_code=404, detail="video export job does not exist")
+        return state
+
+    @router.get("/export-jobs/{run_id}/files/{file_name}")
+    def export_job_file(run_id: str, file_name: str) -> FileResponse:
+        path = export_manager.file_path(run_id, file_name)
+        if path is None:
+            raise HTTPException(status_code=404, detail="输出文件不存在")
+        media_type = "video/mp4" if file_name.endswith(".mp4") else "application/octet-stream"
+        return FileResponse(path, media_type=media_type, filename=file_name)
+
+    return router
+
+
+def _gpu_info() -> dict | None:
+    try:
+        completed = subprocess.run(
+            ["nvidia-smi", "--query-gpu=name,memory.total,driver_version", "--format=csv,noheader,nounits"],
+            check=True,
+            capture_output=True,
+            text=True,
+            timeout=3,
+        )
+    except Exception:
+        return None
+    first = completed.stdout.strip().splitlines()[0] if completed.stdout.strip() else ""
+    parts = [item.strip() for item in first.split(",")]
+    if len(parts) < 3:
+        return None
+    return {"name": parts[0], "memory_total_mb": int(float(parts[1])), "driver_version": parts[2]}

ai-services/vision-inference/app/video_demo_schemas.py

@@ -0,0 +1,57 @@
+from __future__ import annotations
+
+from typing import Any, Literal
+
+from pydantic import BaseModel, Field
+
+
+VideoExportJobState = Literal["queued", "running", "succeeded", "failed", "cancelled"]
+
+
+class VideoDemoWarning(BaseModel):
+    code: str
+    message: str
+    model_group: str | None = None
+
+
+class WarmupRequest(BaseModel):
+    models: list[str] = Field(default_factory=list)
+    model_version: str | None = None
+    model_versions: dict[str, str] = Field(default_factory=dict)
+
+
+class VideoExportJobCreateResponse(BaseModel):
+    run_id: str
+    status: VideoExportJobState
+    output_dir: str
+    status_url: str
+
+
+class VideoExportProgress(BaseModel):
+    processed_frames: int = 0
+    total_frames: int | None = None
+    percent: float = 0.0
+    elapsed_seconds: float = 0.0
+    eta_seconds: float | None = None
+
+
+class VideoExportOutputs(BaseModel):
+    annotated_video: str | None = None
+    results_json: str | None = None
+    results_jsonl: str | None = None
+    metadata_json: str | None = None
+    log: str | None = None
+
+
+class VideoExportJobStatus(BaseModel):
+    run_id: str
+    status: VideoExportJobState
+    progress: VideoExportProgress
+    outputs: VideoExportOutputs
+    warnings: list[VideoDemoWarning] = Field(default_factory=list)
+    error: str | None = None
+    cancel_requested: bool = False
+    cancelled_at: str | None = None
+
+
+JsonMap = dict[str, Any]

ai-services/vision-inference/app/video_export_runtime.py

@@ -0,0 +1,671 @@
+from __future__ import annotations
+
+import json
+import os
+import re
+import threading
+import time
+import uuid
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+from urllib.parse import unquote
+
+import cv2
+import numpy as np
+from fastapi import UploadFile
+
+from .cpu_runtime import CpuVisionRuntime
+from .frame_runtime import FrameInferenceService
+
+
+ALLOWED_OUTPUT_FILES = {"annotated.mp4", "results.json", "results.jsonl", "run-metadata.json", "run.log"}
+DEFAULT_MAX_VIDEO_UPLOAD_BYTES = 2 * 1024 * 1024 * 1024
+VIDEO_UPLOAD_LIMIT_ENV = "RAIL_VIDEO_DEMO_MAX_VIDEO_BYTES"
+VIDEO_MIME_TYPES_BY_SUFFIX = {
+    ".mp4": {"video/mp4", "application/mp4"},
+    ".m4v": {"video/mp4", "video/x-m4v"},
+    ".mov": {"video/quicktime"},
+    ".avi": {"video/x-msvideo", "video/avi"},
+    ".mkv": {"video/x-matroska", "video/mkv"},
+    ".webm": {"video/webm"},
+}
+SAFE_RUN_ID = re.compile(r"[A-Za-z0-9][A-Za-z0-9._-]{0,127}\Z")
+
+
+class VideoUploadError(ValueError):
+    def __init__(self, message: str, status_code: int = 400):
+        super().__init__(message)
+        self.status_code = status_code
+
+
+class VideoExportStateError(RuntimeError):
+    pass
+
+
+class VideoExportCancelled(RuntimeError):
+    pass
+
+
+def max_video_upload_bytes() -> int:
+    configured = os.getenv(VIDEO_UPLOAD_LIMIT_ENV, "").strip()
+    if not configured:
+        return DEFAULT_MAX_VIDEO_UPLOAD_BYTES
+    try:
+        value = int(configured)
+    except ValueError:
+        return DEFAULT_MAX_VIDEO_UPLOAD_BYTES
+    return value if value > 0 else DEFAULT_MAX_VIDEO_UPLOAD_BYTES
+
+
+def _repo_root() -> Path:
+    current = Path(__file__).resolve()
+    for parent in current.parents:
+        if (parent / "visualization-demo").is_dir():
+            return parent
+    return Path.cwd()
+
+
+def _output_root() -> Path:
+    configured = os.getenv("RAIL_VIDEO_DEMO_OUTPUT_DIR")
+    if configured:
+        return Path(configured).resolve()
+    return (_repo_root() / "visualization-demo" / "outputs" / "video-runs").resolve()
+
+
+class VideoExportManager:
+    def __init__(self, runtime: CpuVisionRuntime, output_root: Path | None = None):
+        self.runtime = runtime
+        self.frame_service = FrameInferenceService(runtime)
+        self.output_root = (output_root or _output_root()).resolve()
+        self.output_root.mkdir(parents=True, exist_ok=True)
+        self._lock = threading.RLock()
+        self._jobs: dict[str, dict[str, Any]] = {}
+        self._cancel_events: dict[str, threading.Event] = {}
+
+    async def create_job(
+        self,
+        upload: UploadFile,
+        *,
+        detect_enabled: bool,
+        segment_enabled: bool,
+        confidence_threshold: float,
+        mask_threshold: float,
+        max_inference_width: int,
+        analysis_stride: int,
+        reuse_last_result: bool,
+        max_detections: int,
+        detection_scene: str,
+        detection_model_version: str | None,
+        segmentation_model_version: str | None,
+    ) -> dict[str, Any]:
+        source_suffix = self._validate_upload(upload)
+        upload_limit = max_video_upload_bytes()
+        if upload.size is not None and upload.size > upload_limit:
+            raise VideoUploadError(
+                f"video upload exceeds the configured limit of {upload_limit} bytes",
+                status_code=413,
+            )
+
+        run_id = self._new_run_id()
+        run_dir = self._run_dir(run_id)
+        if run_dir is None:
+            raise RuntimeError("generated video export run id is invalid")
+        run_dir.mkdir(parents=True, exist_ok=False)
+
+        source_path = run_dir / f"source{source_suffix}"
+        written = 0
+        try:
+            with source_path.open("wb") as target:
+                while chunk := await upload.read(1024 * 1024):
+                    written += len(chunk)
+                    if written > upload_limit:
+                        raise VideoUploadError(
+                            f"video upload exceeds the configured limit of {upload_limit} bytes",
+                            status_code=413,
+                        )
+                    target.write(chunk)
+            if written == 0:
+                raise VideoUploadError("video upload is empty")
+        except Exception:
+            source_path.unlink(missing_ok=True)
+            try:
+                run_dir.rmdir()
+            except OSError:
+                pass
+            raise
+
+        config = {
+            "detect_enabled": detect_enabled,
+            "segment_enabled": segment_enabled,
+            "confidence_threshold": confidence_threshold,
+            "mask_threshold": mask_threshold,
+            "max_inference_width": max_inference_width,
+            "analysis_stride": max(1, analysis_stride),
+            "reuse_last_result": reuse_last_result,
+            "max_detections": max(1, max_detections),
+            "detection_scene": detection_scene,
+            "detection_model_version": detection_model_version,
+            "segmentation_model_version": segmentation_model_version,
+            "source_path": str(source_path),
+            "source_name": upload.filename or source_path.name,
+            "source_size_bytes": written,
+        }
+        state = self._state(run_id, "queued", run_dir)
+        self._save_json(run_dir / "status.json", state)
+        self._save_json(run_dir / "run-metadata.json", {"run_id": run_id, "created_at": self._now(), "config": config})
+        with self._lock:
+            self._jobs[run_id] = {**state, "config": config}
+            self._cancel_events[run_id] = threading.Event()
+        return {"run_id": run_id, "status": "queued", "output_dir": str(run_dir)}
+
+    def run_job(self, run_id: str) -> None:
+        run_dir = self._run_dir(run_id)
+        if run_dir is None:
+            return
+        with self._lock:
+            job = self._jobs.get(run_id)
+        if not job:
+            return
+
+        config = dict(job["config"])
+        log_path = run_dir / "run.log"
+        source_path = Path(config["source_path"])
+        annotated_path = run_dir / "annotated.mp4"
+        results_jsonl_path = run_dir / "results.jsonl"
+        results_json_path = run_dir / "results.json"
+        metadata_path = run_dir / "run-metadata.json"
+
+        started = time.perf_counter()
+        warnings: list[dict[str, Any]] = []
+        frame_records: list[dict[str, Any]] = []
+        frame_index = 0
+        total_frames: int | None = None
+        metadata: dict[str, Any] = {}
+        capture: cv2.VideoCapture | None = None
+        writer: cv2.VideoWriter | None = None
+        try:
+            self._raise_if_cancelled(run_id)
+            running_state = self._update(run_id, "running", processed_frames=0, total_frames=None, warnings=warnings)
+            if running_state["status"] != "running":
+                raise VideoExportCancelled()
+            self._append_log(log_path, f"started source={source_path}")
+
+            capture = cv2.VideoCapture(str(source_path))
+            if not capture.isOpened():
+                raise RuntimeError(f"无法打开视频文件: {source_path}")
+
+            width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH) or 0)
+            height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT) or 0)
+            fps = float(capture.get(cv2.CAP_PROP_FPS) or 0) or 25.0
+            total_frames_raw = int(capture.get(cv2.CAP_PROP_FRAME_COUNT) or 0)
+            total_frames = total_frames_raw if total_frames_raw > 0 else None
+            if width <= 0 or height <= 0:
+                raise RuntimeError("视频宽高无效，无法生成结果视频")
+
+            fourcc = cv2.VideoWriter_fourcc(*"mp4v")
+            writer = cv2.VideoWriter(str(annotated_path), fourcc, fps, (width, height))
+            if not writer.isOpened():
+                raise RuntimeError("OpenCV VideoWriter 无法创建 annotated.mp4")
+
+            metadata = {
+                "run_id": run_id,
+                "created_at": self._now(),
+                "started_at": self._now(),
+                "source": {"path": str(source_path), "name": config["source_name"], "width": width, "height": height, "fps": fps, "frames": total_frames},
+                "output": {"path": str(annotated_path), "codec": "mp4v"},
+                "runtime": {"provider": self.runtime.settings.execution_provider},
+                "config": config,
+            }
+            self._save_json(metadata_path, metadata)
+
+            last_inference: dict[str, Any] | None = None
+            warning_keys: set[tuple[str, str, str]] = set()
+            with results_jsonl_path.open("w", encoding="utf-8") as stream:
+                while True:
+                    self._raise_if_cancelled(run_id)
+                    ok, frame = capture.read()
+                    if not ok:
+                        break
+
+                    should_infer = frame_index % int(config["analysis_stride"]) == 0
+                    if should_infer or not config["reuse_last_result"] or last_inference is None:
+                        inference = self.frame_service.infer_frame(
+                            frame,
+                            detect_enabled=bool(config["detect_enabled"]),
+                            segment_enabled=bool(config["segment_enabled"]),
+                            confidence_threshold=float(config["confidence_threshold"]),
+                            mask_threshold=float(config["mask_threshold"]),
+                            max_detections=int(config["max_detections"]),
+                            max_inference_width=int(config["max_inference_width"]),
+                            detection_scene=str(config.get("detection_scene") or "inspection"),
+                            detection_model_version=config.get("detection_model_version"),
+                            segmentation_model_version=config.get("segmentation_model_version"),
+                        )
+                        last_inference = inference
+                        self._merge_warnings(warnings, warning_keys, inference.get("warnings", []))
+                    else:
+                        inference = last_inference
+
+                    # Model inference is the longest non-interruptible operation.
+                    # Check again before writing so cancellation stops on this frame.
+                    self._raise_if_cancelled(run_id)
+
+                    timestamp_ms = round(frame_index * 1000 / fps, 2)
+                    record = {
+                        "frame_index": frame_index,
+                        "timestamp_ms": timestamp_ms,
+                        "inferred": should_infer,
+                        "results": inference["results"],
+                        "runtime": inference["runtime"],
+                        "warnings": inference.get("warnings", []),
+                    }
+                    stream.write(json.dumps(record, ensure_ascii=False) + "\n")
+                    if len(frame_records) < 5000:
+                        frame_records.append(record)
+
+                    writer.write(self._overlay(frame, inference, frame_index, timestamp_ms))
+                    frame_index += 1
+
+                    if frame_index % 10 == 0:
+                        elapsed = time.perf_counter() - started
+                        eta = self._eta(elapsed, frame_index, total_frames)
+                        self._update(
+                            run_id,
+                            "running",
+                            processed_frames=frame_index,
+                            total_frames=total_frames,
+                            elapsed_seconds=elapsed,
+                            eta_seconds=eta,
+                            warnings=warnings[-20:],
+                        )
+
+            self._raise_if_cancelled(run_id)
+            elapsed = time.perf_counter() - started
+            summary = {
+                "run_id": run_id,
+                "status": "succeeded",
+                "source": metadata["source"],
+                "outputs": {
+                    "annotated_video": str(annotated_path),
+                    "results_json": str(results_json_path),
+                    "results_jsonl": str(results_jsonl_path),
+                    "metadata_json": str(metadata_path),
+                    "log": str(log_path),
+                },
+                "processed_frames": frame_index,
+                "elapsed_seconds": round(elapsed, 2),
+                "sample_records": frame_records,
+                "warnings": warnings[-100:],
+            }
+            self._save_json(results_json_path, summary)
+            metadata["finished_at"] = self._now()
+            metadata["elapsed_seconds"] = round(elapsed, 2)
+            self._save_json(metadata_path, metadata)
+            if writer is not None:
+                writer.release()
+                writer = None
+            final_state = self._update(
+                run_id,
+                "succeeded",
+                processed_frames=frame_index,
+                total_frames=total_frames or frame_index,
+                elapsed_seconds=elapsed,
+                eta_seconds=0,
+                warnings=warnings[-20:],
+            )
+            if final_state["status"] != "succeeded":
+                raise VideoExportCancelled()
+            self._append_log(log_path, f"succeeded frames={frame_index} elapsed={elapsed:.2f}s")
+        except VideoExportCancelled:
+            elapsed = time.perf_counter() - started
+            if writer is not None:
+                writer.release()
+                writer = None
+            if capture is not None:
+                capture.release()
+                capture = None
+            summary = {
+                "run_id": run_id,
+                "status": "cancelled",
+                "source": metadata.get("source", {"path": str(source_path), "name": config["source_name"]}),
+                "outputs": {
+                    "annotated_video": str(annotated_path) if annotated_path.is_file() else None,
+                    "results_json": str(results_json_path),
+                    "results_jsonl": str(results_jsonl_path) if results_jsonl_path.is_file() else None,
+                    "metadata_json": str(metadata_path),
+                    "log": str(log_path),
+                },
+                "processed_frames": frame_index,
+                "elapsed_seconds": round(elapsed, 2),
+                "sample_records": frame_records,
+                "warnings": warnings[-100:],
+            }
+            self._save_json(results_json_path, summary)
+            if not metadata:
+                metadata = {
+                    "run_id": run_id,
+                    "created_at": self._now(),
+                    "source": {"path": str(source_path), "name": config["source_name"]},
+                    "config": config,
+                }
+            metadata["cancelled_at"] = self._now()
+            metadata["elapsed_seconds"] = round(elapsed, 2)
+            self._save_json(metadata_path, metadata)
+            self._update(
+                run_id,
+                "cancelled",
+                processed_frames=frame_index,
+                total_frames=total_frames,
+                elapsed_seconds=elapsed,
+                eta_seconds=None,
+                warnings=warnings[-20:],
+                cancel_requested=True,
+                cancelled_at=metadata["cancelled_at"],
+            )
+            self._append_log(log_path, f"cancelled frames={frame_index} elapsed={elapsed:.2f}s")
+        except Exception as exc:
+            elapsed = time.perf_counter() - started
+            failed_state = self._update(
+                run_id,
+                "failed",
+                elapsed_seconds=elapsed,
+                error=f"{type(exc).__name__}: {exc}",
+                warnings=warnings[-20:],
+            )
+            if failed_state["status"] == "cancelled":
+                self._append_log(log_path, f"stopped after cancellation {type(exc).__name__}: {exc}")
+            else:
+                self._append_log(log_path, f"failed {type(exc).__name__}: {exc}")
+        finally:
+            if capture is not None:
+                capture.release()
+            if writer is not None:
+                writer.release()
+
+    def status(self, run_id: str) -> dict[str, Any] | None:
+        run_dir = self._run_dir(run_id)
+        if run_dir is None:
+            return None
+        with self._lock:
+            state = self._jobs.get(run_id)
+        if state:
+            return self._public_state(state)
+        path = run_dir / "status.json"
+        if path.is_file():
+            return json.loads(path.read_text(encoding="utf-8"))
+        return None
+
+    def cancel(self, run_id: str) -> dict[str, Any] | None:
+        run_dir = self._run_dir(run_id)
+        if run_dir is None:
+            return None
+        with self._lock:
+            state = self._jobs.get(run_id)
+            if state is None:
+                status_path = run_dir / "status.json"
+                if not status_path.is_file():
+                    return None
+                state = json.loads(status_path.read_text(encoding="utf-8"))
+
+            current_status = str(state.get("status", ""))
+            if current_status in {"succeeded", "failed"}:
+                raise VideoExportStateError(f"video export job is already {current_status}")
+            if current_status == "cancelled":
+                return self._public_state(state)
+            if current_status not in {"queued", "running"}:
+                raise VideoExportStateError(f"video export job cannot be cancelled from {current_status or 'unknown'}")
+
+            event = self._cancel_events.setdefault(run_id, threading.Event())
+            event.set()
+            progress = dict(state.get("progress") or {})
+            cancelled_at = self._now()
+            cancelled = self._state(
+                run_id,
+                "cancelled",
+                run_dir,
+                processed_frames=int(progress.get("processed_frames") or 0),
+                total_frames=progress.get("total_frames"),
+                elapsed_seconds=float(progress.get("elapsed_seconds") or 0),
+                eta_seconds=None,
+                warnings=list(state.get("warnings") or []),
+                cancel_requested=True,
+                cancelled_at=cancelled_at,
+            )
+            config = state.get("config")
+            self._jobs[run_id] = {**cancelled, **({"config": config} if config else {})}
+            self._save_json(run_dir / "status.json", cancelled)
+
+        self._append_log(run_dir / "run.log", f"cancel requested previous_status={current_status}")
+        return cancelled
+
+    def file_path(self, run_id: str, file_name: str) -> Path | None:
+        if file_name not in ALLOWED_OUTPUT_FILES:
+            return None
+        run_dir = self._run_dir(run_id)
+        if run_dir is None:
+            return None
+        path = (run_dir / file_name).resolve()
+        try:
+            self._ensure_inside_root(path)
+        except ValueError:
+            return None
+        return path if path.is_file() else None
+
+    def _overlay(self, frame: np.ndarray, inference: dict[str, Any], frame_index: int, timestamp_ms: float) -> np.ndarray:
+        output = frame.copy()
+        overlay = output.copy()
+        height, width = output.shape[:2]
+
+        for segment in inference["results"].get("segments", []):
+            color = self._color(str(segment.get("category", "segment")))
+            mask = self._decode_mask(segment.get("mask"))
+            if mask is not None:
+                scaled = cv2.resize(mask, (width, height), interpolation=cv2.INTER_NEAREST)
+                active = scaled > 0
+                overlay[active] = color
+                contours, _ = cv2.findContours((active.astype(np.uint8) * 255), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+                if contours:
+                    cv2.drawContours(output, contours, -1, color, 2, cv2.LINE_AA)
+                continue
+            polygon = segment.get("polygon") or []
+            if len(polygon) < 3:
+                continue
+            points = np.asarray(
+                [[int(self._clip(point[0]) * width), int(self._clip(point[1]) * height)] for point in polygon],
+                dtype=np.int32,
+            )
+            cv2.fillPoly(overlay, [points], color)
+            cv2.polylines(output, [points], True, color, 2, cv2.LINE_AA)
+        output = cv2.addWeighted(overlay, 0.35, output, 0.65, 0)
+
+        for detection in inference["results"].get("detections", []):
+            bbox = detection.get("bbox") or []
+            if len(bbox) != 4:
+                continue
+            x1, y1, x2, y2 = bbox
+            left = int(self._clip(x1) * width)
+            top = int(self._clip(y1) * height)
+            right = int(self._clip(x2) * width)
+            bottom = int(self._clip(y2) * height)
+            color = self._color(str(detection.get("category", "target")))
+            cv2.rectangle(output, (left, top), (right, bottom), color, 2)
+            label = f"{detection.get('category', 'target')} {float(detection.get('confidence', 0)):.2f}"
+            cv2.putText(output, label, (left, max(18, top - 6)), cv2.FONT_HERSHEY_SIMPLEX, 0.55, color, 2, cv2.LINE_AA)
+
+        cv2.putText(output, f"frame {frame_index}  {timestamp_ms / 1000:.2f}s", (12, 28), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
+        cv2.putText(output, "Demo - not production alarm evidence", (12, height - 16), cv2.FONT_HERSHEY_SIMPLEX, 0.55, (255, 255, 255), 2, cv2.LINE_AA)
+        return output
+
+    def _state(
+        self,
+        run_id: str,
+        status: str,
+        run_dir: Path,
+        *,
+        processed_frames: int = 0,
+        total_frames: int | None = None,
+        elapsed_seconds: float = 0.0,
+        eta_seconds: float | None = None,
+        warnings: list[dict[str, Any]] | None = None,
+        error: str | None = None,
+        cancel_requested: bool = False,
+        cancelled_at: str | None = None,
+    ) -> dict[str, Any]:
+        percent = round(processed_frames * 100 / total_frames, 2) if total_frames else 0.0
+        return {
+            "run_id": run_id,
+            "status": status,
+            "progress": {
+                "processed_frames": processed_frames,
+                "total_frames": total_frames,
+                "percent": percent,
+                "elapsed_seconds": round(elapsed_seconds, 2),
+                "eta_seconds": round(eta_seconds, 2) if eta_seconds is not None else None,
+            },
+            "outputs": {
+                "annotated_video": str(run_dir / "annotated.mp4") if (run_dir / "annotated.mp4").is_file() else None,
+                "results_json": str(run_dir / "results.json") if (run_dir / "results.json").is_file() else None,
+                "results_jsonl": str(run_dir / "results.jsonl") if (run_dir / "results.jsonl").is_file() else None,
+                "metadata_json": str(run_dir / "run-metadata.json") if (run_dir / "run-metadata.json").is_file() else None,
+                "log": str(run_dir / "run.log"),
+            },
+            "warnings": warnings or [],
+            "error": error,
+            "cancel_requested": cancel_requested,
+            "cancelled_at": cancelled_at,
+        }
+
+    def _update(self, run_id: str, status: str, **kwargs: Any) -> dict[str, Any]:
+        run_dir = self._run_dir(run_id)
+        if run_dir is None:
+            raise ValueError(f"invalid video export run id: {run_id}")
+        with self._lock:
+            previous = self._jobs.get(run_id, {})
+            if previous.get("status") == "cancelled" and status != "cancelled":
+                return self._public_state(previous)
+            config = previous.get("config")
+            state = self._state(run_id, status, run_dir, **kwargs)
+            self._jobs[run_id] = {**state, **({"config": config} if config else {})}
+            self._save_json(run_dir / "status.json", state)
+        return state
+
+    def _public_state(self, state: dict[str, Any]) -> dict[str, Any]:
+        return {key: value for key, value in state.items() if key != "config"}
+
+    def _new_run_id(self) -> str:
+        return f"video-run-{datetime.now().strftime('%Y%m%d-%H%M%S')}-{uuid.uuid4().hex[:6]}"
+
+    def _validate_upload(self, upload: UploadFile) -> str:
+        filename = (upload.filename or "").strip()
+        decoded_name = unquote(filename)
+        if not decoded_name:
+            raise VideoUploadError("video filename is required")
+        if "\x00" in decoded_name or "/" in decoded_name or "\\" in decoded_name or decoded_name in {".", ".."}:
+            raise VideoUploadError("video filename must not contain a path")
+        suffix = Path(decoded_name).suffix.lower()
+        allowed_mime_types = VIDEO_MIME_TYPES_BY_SUFFIX.get(suffix)
+        if allowed_mime_types is None:
+            supported = ", ".join(sorted(VIDEO_MIME_TYPES_BY_SUFFIX))
+            raise VideoUploadError(f"unsupported video extension; supported extensions: {supported}")
+        content_type = (upload.content_type or "").split(";", 1)[0].strip().lower()
+        if content_type not in allowed_mime_types:
+            raise VideoUploadError(
+                f"content type {content_type or '<missing>'} does not match {suffix}",
+                status_code=415,
+            )
+        return suffix
+
+    def _run_dir(self, run_id: str) -> Path | None:
+        decoded = unquote(str(run_id))
+        if decoded != run_id or not SAFE_RUN_ID.fullmatch(decoded) or decoded in {".", ".."}:
+            return None
+        path = (self.output_root / decoded).resolve()
+        try:
+            self._ensure_inside_root(path)
+        except ValueError:
+            return None
+        return path
+
+    def _raise_if_cancelled(self, run_id: str) -> None:
+        with self._lock:
+            event = self._cancel_events.get(run_id)
+            state = self._jobs.get(run_id)
+            cancelled = bool(event and event.is_set()) or bool(state and state.get("status") == "cancelled")
+        if cancelled:
+            raise VideoExportCancelled()
+
+    def _ensure_inside_root(self, path: Path) -> None:
+        root = self.output_root.resolve()
+        resolved = path.resolve()
+        if root != resolved and root not in resolved.parents:
+            raise ValueError(f"输出路径越界: {resolved}")
+
+    @staticmethod
+    def _decode_mask(mask: Any) -> np.ndarray | None:
+        if not isinstance(mask, dict) or mask.get("encoding") != "rle":
+            return None
+        try:
+            width = int(mask.get("width", 0))
+            height = int(mask.get("height", 0))
+        except (TypeError, ValueError):
+            return None
+        counts = mask.get("counts")
+        if width <= 0 or height <= 0 or not isinstance(counts, list):
+            return None
+        total = width * height
+        flat = np.zeros(total, dtype=np.uint8)
+        offset = 0
+        value = 0
+        for raw_run in counts:
+            try:
+                run_length = int(raw_run)
+            except (TypeError, ValueError):
+                return None
+            if run_length < 0:
+                return None
+            end = min(total, offset + run_length)
+            if value:
+                flat[offset:end] = 1
+            offset = end
+            if offset >= total:
+                break
+            value = 1 - value
+        return flat.reshape((height, width))
+
+    @staticmethod
+    def _save_json(path: Path, payload: dict[str, Any]) -> None:
+        path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    @staticmethod
+    def _append_log(path: Path, message: str) -> None:
+        with path.open("a", encoding="utf-8") as stream:
+            stream.write(f"{datetime.now().isoformat(timespec='seconds')} {message}\n")
+
+    @staticmethod
+    def _eta(elapsed: float, processed: int, total: int | None) -> float | None:
+        if not total or processed <= 0:
+            return None
+        return max(0.0, elapsed / processed * (total - processed))
+
+    @staticmethod
+    def _merge_warnings(target: list[dict[str, Any]], keys: set[tuple[str, str, str]], incoming: list[dict[str, Any]]) -> None:
+        for item in incoming:
+            key = (str(item.get("code", "")), str(item.get("model_group", "")), str(item.get("message", "")))
+            if key in keys:
+                continue
+            keys.add(key)
+            target.append(item)
+
+    @staticmethod
+    def _now() -> str:
+        return datetime.now().isoformat(timespec="seconds")
+
+    @staticmethod
+    def _clip(value: float) -> float:
+        return min(1.0, max(0.0, float(value)))
+
+    @staticmethod
+    def _color(label: str) -> tuple[int, int, int]:
+        seed = abs(hash(label))
+        return 64 + seed % 160, 64 + (seed // 7) % 160, 64 + (seed // 13) % 160

ai-services/vision-inference/requirements-server.txt

@@ -1,7 +1,8 @@
 fastapi==0.115.5
 uvicorn[standard]==0.32.1
 pydantic==2.10.2
+python-multipart==0.0.19
 opencv-python-headless==4.10.0.84
 numpy==2.1.3
-onnxruntime-gpu==1.20.1
+onnxruntime-gpu[cuda,cudnn]==1.22.0
 minio==7.2.12

ai-services/vision-inference/requirements.txt

@@ -1,6 +1,7 @@
 fastapi==0.115.5
 uvicorn[standard]==0.32.1
 pydantic==2.10.2
+python-multipart==0.0.19
 opencv-python-headless==4.10.0.84
 numpy==2.1.3
 onnxruntime==1.20.1

ai-services/vision-inference/tests/test_inference.py

@@ -51,10 +51,12 @@ def test_cpu_runtime_reports_registered_visual_models():
     assert body["profile"] == "cpu-local"
     assert body["runtime_available"] is True
     assert body["max_concurrency"] == 1
-    assert body["max_loaded_models"] == 1
+    assert body["max_loaded_models"] == 2
     groups = {item["model_group"] for item in body["models"]}
     assert {"vision-detector", "vision-segmenter", "thermal-analyzer"}.issubset(groups)
-    assert {item["model_version"] for item in body["models"]} == {"cpu-v1.0.0"}
+    versions = {item["model_version"] for item in body["models"]}
+    assert "cpu-v1.0.0" in versions
+    assert "traffic-yolov8n-coco" in versions
     assert body["execution_provider_ready"] is True
 
 

ai-services/vision-inference/tests/test_video_demo.py

@@ -0,0 +1,296 @@
+import asyncio
+import io
+import json
+import threading
+import time
+from pathlib import Path
+
+import cv2
+import numpy as np
+import pytest
+from fastapi import FastAPI, UploadFile
+from fastapi.testclient import TestClient
+from starlette.datastructures import Headers
+
+from app.main import app, engine
+from app.cpu_runtime import RuntimeOutcome
+from app.frame_runtime import FrameInferenceService
+from app.video_demo_routes import MAX_FRAME_UPLOAD_BYTES, create_video_demo_router
+from app.video_export_runtime import VideoExportManager
+
+
+def _inference_result() -> dict:
+    return {
+        "results": {"detections": [], "segments": []},
+        "runtime": {"provider": "test", "total_latency_ms": 0.0},
+        "warnings": [],
+    }
+
+
+def _manager(tmp_path: Path) -> VideoExportManager:
+    manager = VideoExportManager(engine.runtime, tmp_path)
+    manager.frame_service.infer_frame = lambda *args, **kwargs: _inference_result()
+    return manager
+
+
+def _client(manager: VideoExportManager) -> TestClient:
+    test_app = FastAPI()
+    test_app.include_router(create_video_demo_router(engine, manager))
+    return TestClient(test_app)
+
+
+def _upload(payload: bytes, filename: str = "sample.mp4", content_type: str = "video/mp4") -> UploadFile:
+    return UploadFile(
+        file=io.BytesIO(payload),
+        size=len(payload),
+        filename=filename,
+        headers=Headers({"content-type": content_type}),
+    )
+
+
+def _create_job(manager: VideoExportManager, payload: bytes, filename: str = "sample.mp4") -> dict:
+    return asyncio.run(
+        manager.create_job(
+            _upload(payload, filename),
+            detect_enabled=True,
+            segment_enabled=False,
+            confidence_threshold=0.45,
+            mask_threshold=0.5,
+            max_inference_width=640,
+            analysis_stride=1,
+            reuse_last_result=True,
+            max_detections=10,
+            detection_scene="inspection",
+            detection_model_version=None,
+            segmentation_model_version=None,
+        )
+    )
+
+
+def _video_bytes(path: Path, frame_count: int = 6) -> bytes:
+    writer = cv2.VideoWriter(str(path), cv2.VideoWriter_fourcc(*"mp4v"), 12.0, (64, 48))
+    assert writer.isOpened()
+    for index in range(frame_count):
+        frame = np.full((48, 64, 3), (index * 17) % 255, dtype=np.uint8)
+        cv2.rectangle(frame, (8 + index % 8, 10), (32, 34), (0, 255, 0), -1)
+        writer.write(frame)
+    writer.release()
+    payload = path.read_bytes()
+    assert payload
+    return payload
+
+
+def test_video_demo_capabilities():
+    client = TestClient(app)
+    response = client.get("/api/v1/video-demo/capabilities")
+    assert response.status_code == 200
+    body = response.json()
+    assert "models" in body
+    assert body["recommended"]["max_inference_width"] in {640, 960}
+    assert body["export"]["enabled"] is True
+
+
+def test_video_demo_infer_frame_reports_model_warning_when_artifact_missing():
+    client = TestClient(app)
+    image = np.zeros((32, 48, 3), dtype=np.uint8)
+    ok, encoded = cv2.imencode(".jpg", image)
+    assert ok
+
+    response = client.post(
+        "/api/v1/video-demo/infer-frame",
+        files={"frame": ("frame.jpg", encoded.tobytes(), "image/jpeg")},
+        data={
+            "session_id": "test-session",
+            "timestamp_ms": "120",
+            "detect_enabled": "true",
+            "segment_enabled": "false",
+            "confidence_threshold": "0.45",
+            "mask_threshold": "0.5",
+            "max_detections": "10",
+            "max_inference_width": "640",
+        },
+    )
+
+    assert response.status_code == 200
+    body = response.json()
+    assert body["session_id"] == "test-session"
+    assert body["source"]["width"] == 48
+    assert body["source"]["height"] == 32
+    assert body["results"]["detections"][0]["model_group"] == "opencv-demo-detector"
+    assert body["warnings"][0]["code"] == "DEMO_FALLBACK_ACTIVE"
+    assert body["warnings"][0]["model_group"] == "vision-detector"
+
+
+def test_frame_segments_preserve_pixel_mask_rle():
+    service = FrameInferenceService(engine.runtime)
+    mask = {"encoding": "rle", "width": 2, "height": 2, "counts": [1, 2, 1]}
+    outcome = RuntimeOutcome(
+        "onnxruntime-cpu",
+        [
+            {
+                "category": "road",
+                "confidence": 0.93,
+                "geometry": {"type": "Polygon", "coordinates": [[]], "coordinate_space": "normalized"},
+                "measurements": {"area_ratio": 0.5},
+                "mask": mask,
+            }
+        ],
+        None,
+        3.4,
+        {"model_group": "vision-segmenter", "model_version": "test-segmenter"},
+    )
+
+    segments = service._segments(outcome)
+
+    assert segments[0]["mask"] == mask
+    assert segments[0]["polygon"] == []
+
+
+def test_video_export_overlay_decodes_pixel_mask_rle(tmp_path):
+    manager = _manager(tmp_path / "runs")
+    frame = np.zeros((80, 80, 3), dtype=np.uint8)
+    mask = {"encoding": "rle", "width": 4, "height": 4, "counts": [5, 2, 2, 2, 5]}
+    inference = {
+        "results": {
+            "segments": [{"category": "road", "confidence": 1.0, "polygon": [], "mask": mask}],
+            "detections": [],
+        }
+    }
+
+    decoded = manager._decode_mask(mask)
+    output = manager._overlay(frame, inference, 0, 0)
+
+    assert decoded is not None
+    assert int(decoded.sum()) == 4
+    assert int(output[40, 40].sum()) > 0
+
+
+def test_video_export_job_requires_at_least_one_enabled_model():
+    client = TestClient(app)
+    response = client.post(
+        "/api/v1/video-demo/export-jobs",
+        files={"video": ("sample.mp4", b"not-a-real-video", "video/mp4")},
+        data={"detect_enabled": "false", "segment_enabled": "false"},
+    )
+    assert response.status_code == 400
+
+
+def test_infer_frame_rejects_payload_over_two_megabytes():
+    client = TestClient(app)
+    response = client.post(
+        "/api/v1/video-demo/infer-frame",
+        files={"frame": ("frame.jpg", b"x" * (MAX_FRAME_UPLOAD_BYTES + 1), "image/jpeg")},
+    )
+    assert response.status_code == 413
+
+
+@pytest.mark.parametrize(
+    ("filename", "content_type", "expected_status"),
+    [
+        ("sample.txt", "video/mp4", 400),
+        ("sample.mp4", "text/plain", 415),
+        ("../escape.mp4", "video/mp4", 400),
+        ("..%2Fescape.mp4", "video/mp4", 400),
+    ],
+)
+def test_video_export_rejects_invalid_extension_mime_and_paths(tmp_path, filename, content_type, expected_status):
+    manager = _manager(tmp_path / "runs")
+    response = _client(manager).post(
+        "/api/v1/video-demo/export-jobs",
+        files={"video": (filename, b"not-a-video", content_type)},
+    )
+    assert response.status_code == expected_status
+    assert not list(manager.output_root.iterdir())
+
+
+def test_video_export_limit_is_configurable_and_cleans_partial_upload(tmp_path, monkeypatch):
+    manager = _manager(tmp_path / "runs")
+    monkeypatch.setenv("RAIL_VIDEO_DEMO_MAX_VIDEO_BYTES", "8")
+    response = _client(manager).post(
+        "/api/v1/video-demo/export-jobs",
+        files={"video": ("sample.mp4", b"123456789", "video/mp4")},
+    )
+    assert response.status_code == 413
+    assert not list(manager.output_root.iterdir())
+
+
+def test_video_export_path_lookup_blocks_traversal(tmp_path):
+    manager = _manager(tmp_path / "runs")
+    outside = tmp_path / "outside"
+    outside.mkdir()
+    (outside / "run.log").write_text("secret", encoding="utf-8")
+
+    assert manager.status("../outside") is None
+    assert manager.status("..%2Foutside") is None
+    assert manager.file_path("../outside", "run.log") is None
+    assert manager.file_path("safe-run", "../run.log") is None
+
+
+def test_queued_video_export_can_be_cancelled_through_api(tmp_path):
+    manager = _manager(tmp_path / "runs")
+    job = _create_job(manager, b"queued-placeholder")
+    client = _client(manager)
+
+    response = client.post(f"/api/v1/video-demo/export-jobs/{job['run_id']}/cancel")
+    assert response.status_code == 200
+    assert response.json()["status"] == "cancelled"
+    assert response.json()["cancel_requested"] is True
+
+    manager.run_job(job["run_id"])
+    state = manager.status(job["run_id"])
+    assert state is not None
+    assert state["status"] == "cancelled"
+    log = Path(state["outputs"]["log"]).read_text(encoding="utf-8")
+    assert "cancel requested previous_status=queued" in log
+    assert "cancelled" in log
+
+
+def test_running_video_export_stops_promptly_when_cancelled(tmp_path):
+    manager = _manager(tmp_path / "runs")
+    payload = _video_bytes(tmp_path / "running-source.mp4", frame_count=80)
+    job = _create_job(manager, payload)
+    inference_started = threading.Event()
+
+    def slow_inference(*args, **kwargs):
+        inference_started.set()
+        time.sleep(0.05)
+        return _inference_result()
+
+    manager.frame_service.infer_frame = slow_inference
+    worker = threading.Thread(target=manager.run_job, args=(job["run_id"],), daemon=True)
+    worker.start()
+    assert inference_started.wait(timeout=3)
+
+    cancelled = manager.cancel(job["run_id"])
+    assert cancelled is not None
+    assert cancelled["status"] == "cancelled"
+    worker.join(timeout=3)
+    assert not worker.is_alive()
+
+    state = manager.status(job["run_id"])
+    assert state is not None
+    assert state["status"] == "cancelled"
+    assert state["progress"]["processed_frames"] < 80
+    assert Path(state["outputs"]["results_json"]).is_file()
+    assert "cancelled" in Path(state["outputs"]["log"]).read_text(encoding="utf-8")
+
+
+def test_video_export_succeeds_and_publishes_expected_files(tmp_path):
+    manager = _manager(tmp_path / "runs")
+    payload = _video_bytes(tmp_path / "success-source.mp4", frame_count=5)
+    job = _create_job(manager, payload)
+
+    manager.run_job(job["run_id"])
+    state = manager.status(job["run_id"])
+    assert state is not None
+    assert state["status"] == "succeeded"
+    assert state["progress"]["processed_frames"] == 5
+    assert state["progress"]["percent"] == 100.0
+
+    for name in ("annotated_video", "results_json", "results_jsonl", "metadata_json", "log"):
+        assert state["outputs"][name]
+        assert Path(state["outputs"][name]).is_file()
+    summary = json.loads(Path(state["outputs"]["results_json"]).read_text(encoding="utf-8"))
+    assert summary["status"] == "succeeded"
+    assert summary["processed_frames"] == 5
+    assert manager.file_path(job["run_id"], "annotated.mp4") == Path(state["outputs"]["annotated_video"])

docs/无人机项目合作方案总结.md

@@ -0,0 +1,128 @@
+# 铁路无人机巡检系统：三种合作方案总结
+
+> 依据豆包对话整理，供内部决策和与合作方沟通使用。金额、股权及法律条款仍需结合最终需求、硬件预算和律师意见确认。
+
+## 一、合作背景与共同前提
+
+- 业务模式：面向政府/铁路客户提供无人机巡检设备与软件的租赁服务，收入来自持续租金，而非一次性项目回款。
+- 我方投入：从零开发巡检管理平台和铁路专项 AI 缺陷识别算法，负责技术交付、运维和迭代。
+- 对方投入：政企渠道、项目落地、商务签约，以及大疆无人机等硬件采购资金。
+- 系统价值：对话中给出的完整系统工作量为约 2500–4000 人天，公允市场价约 400–700 万元；80 万元属于本次合作的特惠价，不应被表述为软件的公允市场价值。
+- 硬件变化：放弃纵横无人机及高价激光雷达后，硬件采购范围缩小为大疆体系，初始投入明显下降；后续每增加客户，通常仍需追加设备资金。
+
+## 二、三种方案总览
+
+| 方案 | 核心交易 | 我方收益 | 我方责任/风险 | 适用情形 |
+|---|---|---|---|---|
+| 方案一：一次性买断 | 80 万元一次性交付完整业务包 | 回款最快、金额确定 | 放弃后续授权和经营收益；第三方许可边界和后续维护必须写清 | 希望快速回款、一次性完成业务交接 |
+| 方案二：License 授权 | 按项目/套数持续收取授权费 | 可获得持续收入，软件可复制到多客户 | 需提供约定范围内的基础维护；要防止跨项目复用和欠费 | 看好规模化业务但不愿参股，建议优先 |
+| 方案三：技术入股 | 软件技术、渠道资源和硬件现金共同形成股权 | 共享长期经营利润，收益上限最高 | 面临公司债务、财务监管、业务外流、知识产权和退出风险 | 双方已验证履约能力，准备长期共同经营 |
+
+## 三、方案一：一次性完整交付、80 万元买断
+
+### 合作方式
+
+我方按 80 万元一次性交付完整业务包，包括平台程序、源代码、铁路专项 AI 模型、部署包、配置文档、使用说明及当前版本相关知识产权。完整业务包交付后，合作方可在约定业务范围内自行复制部署和运营；但大疆 SDK、第三方软件、开源组件及其他受第三方许可约束的内容，仅能按其原许可范围使用，不能超出许可范围转让。
+
+### 优点
+
+- 回款确定，现金流最清晰。
+- 不成立合资公司、不参与对方经营，不承担对方的经营负债和项目回款风险。
+- 合同和交付边界相对简单，后续沟通成本低。
+
+### 缺点
+
+- 收益上限锁定在 80 万元，无法分享对方后续多客户租赁业务的增长收益。
+- 交付完整业务包后，我方不再享有后续复制部署和经营分成，80 万元即为本次完整转让的主要收益。
+- 交付后对方仍可能提出免费维护、升级或新硬件适配要求，需要明确这些内容不包含在买断价内。
+
+### 必须写入合同
+
+1. 明确本次为完整业务包交付，包含源代码、模型、部署资料及约定范围内的著作权/使用权转移。
+2. 明确第三方 SDK、开源组件和外部服务的许可边界，不把无法转让的第三方权利写成我方可转让资产。
+3. 明确验收标准、付款节点、质保期、维护边界和新增开发收费标准；买断不等于永久免费维护。
+4. 明确对方可以在约定业务范围内复制部署，但不得将受限的第三方组件独立转售或超出许可范围分发。
+
+## 四、方案二：阶梯式 License 授权租赁
+
+### 合作方式与价格
+
+- 首套 License：40 万元/年（续约阶段可按 4 万元/月）。
+- 第二套 License：30 万元/年（续约阶段可按 3 万元/月）。
+- 第三套及以上：20 万元/套/年（续约阶段可按 2 万元/月）。
+- 首次采购至少签订 1 个完整年度；续约时可选择年付或月付。
+- 一套 License 对应一个独立政府租赁项目，禁止一套授权跨多个项目使用。
+
+### 建议的买断机制
+
+合作开始 24 个月内，对方可以行使单套系统买断选择权，特惠买断价为 80 万元；超过 24 个月，特惠价格失效，买断价格按公允市场价重新评估（对话中参考 400 万元起）。
+
+这里的买断按方案一延后执行：前期先按 License 收取租赁费，租赁费是独立的使用/服务费用，不抵扣 80 万元买断价。买断时仍需另行支付 80 万元，交付当时的最新稳定迭代版本完整业务包，包含源代码、模型、部署资料及约定范围内的知识产权。买断完成后，后续版本升级、算法优化和新机型适配另行计费。
+
+### 授权包含与不包含
+
+**包含：**当期合法使用权、现有版本 Bug 修复、基础远程运维和正常使用保障。
+
+**不包含：**新增功能、算法升级、定制报表、新机型适配、现场实施、驻场服务和大规模数据处理；以上内容按单独开发合同或人天报价结算。
+
+### 优点
+
+- 不参股、不合伙，基本隔离合资公司的债务和治理风险。
+- 可随着项目数量增加持续获得收入，软件的可复制性能够转化为长期收益。
+- 首年年付保证最低回款，后续月付降低合作方资金压力。
+- 可以通过账号、设备或项目绑定控制授权范围，欠费时暂停服务；在买断前，合作方只能使用当期 License，不得提前复制完整业务包。
+
+### 缺点
+
+- 回款分期，依赖对方持续获取并签约政府租赁项目。
+- 需要长期提供合同约定范围内的维护和服务。
+- 授权控制、项目识别、数据隔离和欠费停用机制需要落地，否则容易出现一套授权多项目使用。
+
+## 五、方案三：技术入股合资经营
+
+### 对话中的估值逻辑
+
+- 我方软件技术价值按 80 万元作为基础价值。
+- 对方政企渠道资源暂按 80 万元作为谈判基础，但渠道资源是否能够转化为实际出资，需要由律师按法律和公司登记规则处理。
+- 剩余股权根据双方实际硬件/现金出资比例核算。
+- 对话中形成的初步区间为：我方 35%–45%，对方 55%–65%，避免长期 50:50 僵局。
+
+### 合作分工
+
+- 我方：平台开发、算法、技术交付、运维和迭代。
+- 对方：渠道拓展、投标签约、客户维护、硬件采购和租赁业务落地。
+- 项目租金收入扣除硬件折旧、运营成本、税费和运维成本后，按约定股权或利润分配机制结算。
+
+### 优点
+
+- 双方共同经营，能够分享多区域、多客户租赁业务的长期利润。
+- 对方的渠道能力和我方的软件能力可以形成互补，适合需要持续投入和深度服务的业务。
+
+### 缺点与主要风险
+
+- 渠道资源很难像现金或知识产权一样直接作为法定出资，必须以业绩条件、分期成熟和回购机制约束。
+- 合资公司可能产生硬件采购欠款、人员成本、项目违约赔偿等经营负债。
+- 需要防范资金挪用、虚报成本、项目不入账、体外接单、分红争议和财务信息不透明。
+- 软件一旦直接转入合资公司，后续独立接单、授权和退出都会变得困难。
+
+### 建议的安全结构与合同机制
+
+1. 为体现合作诚意，可以将软件著作权、源代码及完整业务包按评估价值转入合资公司；转让范围、评估价格、交付时点和第三方许可边界必须在知识产权协议中单独列明。
+2. 软件资产转入应与项目落地条件、出资到账和交付验收绑定。若 24 个月内未达到约定的客户落地、签约或回款指标，应触发项目终止、资产处置或知识产权返还/回转机制，而不是默认由我方购买合资公司股权。
+3. 设置合资公司对公账户、月度财务报表、重大采购和对外合同双签机制。
+4. 明确同类政企租赁业务必须进入公司，禁止体外循环；发生侵占公司利益、恶意转移业务等情形时触发违约责任、股权处置和退出机制。
+5. 控制注册资本和双方认缴额度，明确后续增资、亏损承担和清算规则。
+
+## 六、综合判断与建议顺序
+
+### 推荐顺序
+
+1. **首选方案二**：兼顾持续收益、知识产权控制和风险隔离，适合先验证项目和合作关系。
+2. **备选方案一**：如果最看重确定性和快速回款，直接一次性买断，但必须明确完整交付范围和后续义务。
+3. **谨慎方案三**：可以用知识产权转入合资公司的方式体现诚意，但必须绑定项目落地条件、资产返还/回转机制和清算安排；我方不承诺在对方未落地业务时收购合资公司股权。
+
+### 发送给对方前的三个确认点
+
+- 方案一的 80 万元是否确认包含源代码、模型、部署资料和约定范围内的知识产权转移；第三方 SDK/开源组件如何处理？
+- 方案二明确：租赁费不抵扣 80 万元买断价；买断时交付哪个“当前迭代版本”；买断后新增维护、升级和适配如何收费？
+- 方案三的软件知识产权是否转入合资公司；如果项目未落地，如何触发资产返还/回转或清算；双方确认我方不承担收购合资公司股权义务。

frontend/src/components/gis/OperationalGisMap.vue

@@ -55,8 +55,8 @@ function visibleCollection() {
 function addLayers() {
   if (!map || map.getSource("railway-features")) return;
   map.addSource("railway-features", { type: "geojson", data: visibleCollection() as GeoJSON.FeatureCollection });
-  map.addLayer({ id: "object-fill", type: "fill", source: "railway-features", filter: ["all", ["==", ["get", "domain_type"], "inspection_object"], ["==", ["geometry-type"], "Polygon"]], paint: { "fill-color": "#4c9c78", "fill-opacity": 0.16, "fill-outline-color": "#287257" } });
+  map.addLayer({ id: "object-fill", type: "fill", source: "railway-features", filter: ["all", ["==", ["get", "domain_type"], "inspection_object"], ["==", ["geometry-type"], "Polygon"]], paint: { "fill-color": ["case", ["==", ["to-string", ["id"]], props.selectedId], "#16a34a", "#4c9c78"], "fill-opacity": ["case", ["==", ["to-string", ["id"]], props.selectedId], 0.34, 0.16], "fill-outline-color": "#287257" } });
-  map.addLayer({ id: "object-line", type: "line", source: "railway-features", filter: ["all", ["==", ["get", "domain_type"], "inspection_object"], ["==", ["geometry-type"], "LineString"]], paint: { "line-color": "#17365d", "line-width": 5, "line-opacity": 0.82 } });
+  map.addLayer({ id: "object-line", type: "line", source: "railway-features", filter: ["all", ["==", ["get", "domain_type"], "inspection_object"], ["==", ["geometry-type"], "LineString"]], paint: { "line-color": ["case", ["==", ["to-string", ["id"]], props.selectedId], "#16a34a", "#17365d"], "line-width": ["case", ["==", ["to-string", ["id"]], props.selectedId], 8, 5], "line-opacity": 0.82 } });
   map.addLayer({ id: "route-line", type: "line", source: "railway-features", filter: ["==", ["get", "domain_type"], "route"], paint: { "line-color": "#2563eb", "line-width": 3, "line-dasharray": [2, 2] } });
   map.addLayer({ id: "track-line", type: "line", source: "railway-features", filter: ["==", ["get", "domain_type"], "flight_track"], paint: { "line-color": "#16a34a", "line-width": 4 } });
   map.addLayer({ id: "device-point", type: "circle", source: "railway-features", filter: ["==", ["get", "domain_type"], "uav_device"], paint: { "circle-radius": 7, "circle-color": "#0f766e", "circle-stroke-color": "#ffffff", "circle-stroke-width": 2 } });
@@ -66,6 +66,14 @@ function addLayers() {
     const properties = event.features?.[0]?.properties as Row | undefined;
     if (properties) emit("select", properties);
   });
+  for (const layerId of ["object-fill", "object-line"]) {
+    map.on("click", layerId, (event) => {
+      const properties = event.features?.[0]?.properties as Row | undefined;
+      if (properties) emit("select", { ...properties, object_id: properties.object_id || event.features?.[0]?.id });
+    });
+    map.on("mouseenter", layerId, () => { if (map) map.getCanvas().style.cursor = "pointer"; });
+    map.on("mouseleave", layerId, () => { if (map) map.getCanvas().style.cursor = ""; });
+  }
   map.on("mouseenter", "alarm-point", () => { if (map) map.getCanvas().style.cursor = "pointer"; });
   map.on("mouseleave", "alarm-point", () => { if (map) map.getCanvas().style.cursor = ""; });
 }
@@ -79,7 +87,9 @@ function eachCoordinate(value: unknown, callback: (coordinate: [number, number])
 function fitFeatures() {
   if (!map) return;
   const bounds = new LngLatBounds();
-  visibleCollection().features.forEach((feature: Row) => eachCoordinate(feature.geometry?.coordinates, (coordinate) => bounds.extend(coordinate)));
+  const features = visibleCollection().features;
+  const focused = props.selectedId ? features.filter((feature: Row) => String(feature.id || feature.properties?.alarm_id || feature.properties?.object_id) === props.selectedId) : [];
+  (focused.length ? focused : features).forEach((feature: Row) => eachCoordinate(feature.geometry?.coordinates, (coordinate) => bounds.extend(coordinate)));
   if (!bounds.isEmpty()) map.fitBounds(bounds, { padding: 52, maxZoom: 15, duration: 0 });
 }
 
@@ -91,6 +101,14 @@ function updateData(fit = false) {
     map.setPaintProperty("alarm-point", "circle-stroke-color", ["case", ["==", ["to-string", ["id"]], props.selectedId], "#111827", "#ffffff"]);
     map.setPaintProperty("alarm-point", "circle-stroke-width", ["case", ["==", ["to-string", ["id"]], props.selectedId], 4, 2]);
   }
+  if (map.getLayer("object-fill")) {
+    map.setPaintProperty("object-fill", "fill-color", ["case", ["==", ["to-string", ["id"]], props.selectedId], "#16a34a", "#4c9c78"]);
+    map.setPaintProperty("object-fill", "fill-opacity", ["case", ["==", ["to-string", ["id"]], props.selectedId], 0.34, 0.16]);
+  }
+  if (map.getLayer("object-line")) {
+    map.setPaintProperty("object-line", "line-color", ["case", ["==", ["to-string", ["id"]], props.selectedId], "#16a34a", "#17365d"]);
+    map.setPaintProperty("object-line", "line-width", ["case", ["==", ["to-string", ["id"]], props.selectedId], 8, 5]);
+  }
   if (fit) fitFeatures();
 }
 
@@ -103,6 +121,6 @@ onMounted(async () => {
 });
 watch(() => props.featureCollection, () => updateData(true), { deep: true });
 watch(() => props.layers, () => updateData(false), { deep: true });
-watch(() => props.selectedId, () => updateData(false));
+watch(() => props.selectedId, () => updateData(true));
 onBeforeUnmount(() => { map?.remove(); map = null; });
 </script>

frontend/src/navigation/sidebar.ts

@@ -9,6 +9,7 @@ import {
   Files,
   FolderOpened,
   Guide,
+  MagicStick,
   MapLocation,
   Monitor,
   Position,
@@ -51,6 +52,7 @@ export const navigationGroups: NavigationGroup[] = [
     label: "研判",
     items: [
       { path: "/analysis", label: "智能分析", description: "推理与规则", icon: Cpu },
+      { path: "/visualization-demo", label: "视频 AI 演示", description: "上传视频与本地导出", icon: MagicStick },
       { path: "/alarms", label: "告警中心", description: "告警研判", icon: Bell }
     ]
   },

frontend/src/router/index.ts

@@ -19,6 +19,7 @@ const router = createRouter({
         { path: "uav-operations", name: "uav-operations", meta: { title: "无人机运行" }, component: () => import("../views/uav-operations/UavOperationsView.vue") },
         { path: "resources/:resourceId?", name: "resources", meta: { title: "数据资源" }, component: () => import("../views/resources/ResourcesView.vue") },
         { path: "analysis/:entityType?/:entityId?", name: "analysis", meta: { title: "智能分析" }, component: () => import("../views/analysis/AnalysisView.vue") },
+        { path: "visualization-demo", name: "visualization-demo", meta: { title: "视频 AI 演示" }, component: () => import("../views/visualization-demo/VideoAiDemoView.vue") },
         { path: "alarms/:alarmId?", name: "alarms", meta: { title: "告警中心" }, component: () => import("../views/alarms/AlarmsView.vue") },
         { path: "workorders/:workorderId?", name: "workorders", meta: { title: "工单中心" }, component: () => import("../views/workorders/WorkordersView.vue") },
         { path: "algorithm-assets/:entityType?/:entityId?", name: "algorithm-assets", meta: { title: "样本与模型" }, component: () => import("../views/algorithm-assets/AlgorithmAssetsView.vue") },

frontend/src/services/api.ts

@@ -53,8 +53,19 @@ export async function analysisResults() {
   return data.data.results;
 }
 
+export async function createAnalysisJob(payload: {
+  task_id: string;
+  resource_ids: string[];
+  scene_set?: string[];
+  analysis_mode?: string;
+  priority?: string;
+}) {
+  const { data } = await client.post("/analysis/jobs", payload);
+  return data.data;
+}
+
 export async function runAnalysisJob(jobId: string) {
-  const { data } = await client.post(`/analysis/jobs/${jobId}/run`, {});
+  const { data } = await client.post(`/analysis/jobs/${jobId}/run`, {}, { timeout: 10 * 60 * 1000 });
   return data.data;
 }
 
@@ -113,6 +124,11 @@ export async function serviceHealth() {
   return data;
 }
 
+export async function operationsHealth() {
+  const { data } = await client.get("/operations/health", { timeout: 15000 });
+  return data.data;
+}
+
 export async function aiRuntimeStatus() {
   const { data } = await client.get("/operations/ai-runtime");
   return data.data;

frontend/src/services/videoDemoApi.ts

@@ -0,0 +1,109 @@
+import axios from "axios";
+
+const videoDemoBase = (import.meta.env.VITE_VISION_API_BASE || "/vision-api").replace(/\/$/, "");
+const client = axios.create({
+  baseURL: `${videoDemoBase}/api/v1/video-demo`,
+  timeout: 45000
+});
+
+export type VideoDemoWarning = {
+  code: string;
+  message: string;
+  model_group?: string;
+};
+
+export type VideoDetection = {
+  category: string;
+  confidence: number;
+  bbox: number[];
+  model_group: string;
+  model_version?: string;
+  execution_mode?: string;
+};
+
+export type VideoMaskRle = {
+  encoding: "rle";
+  width: number;
+  height: number;
+  counts: number[];
+};
+
+export type VideoSegment = {
+  category: string;
+  confidence: number;
+  polygon: number[][];
+  mask?: VideoMaskRle;
+  area_ratio?: number;
+  model_group: string;
+  model_version?: string;
+  execution_mode?: string;
+};
+
+export type FrameInferenceResponse = {
+  session_id: string;
+  timestamp_ms: number;
+  frame_id: string;
+  source: { width: number; height: number; reported_width?: number; reported_height?: number };
+  inference: { width: number; height: number };
+  scene?: { id: string; label: string; detection_model_version?: string; segmentation_model_version?: string };
+  runtime: Record<string, number | string | undefined>;
+  results: { detections: VideoDetection[]; segments: VideoSegment[] };
+  warnings: VideoDemoWarning[];
+};
+
+export type VideoExportJob = {
+  run_id: string;
+  status: "queued" | "running" | "succeeded" | "failed";
+  output_dir?: string;
+  status_url?: string;
+  progress: {
+    processed_frames: number;
+    total_frames?: number | null;
+    percent: number;
+    elapsed_seconds?: number;
+    eta_seconds?: number | null;
+  };
+  outputs: {
+    annotated_video?: string | null;
+    results_json?: string | null;
+    results_jsonl?: string | null;
+    metadata_json?: string | null;
+    log?: string | null;
+  };
+  warnings: VideoDemoWarning[];
+  error?: string | null;
+};
+
+export type VideoExportCreateResponse = Pick<VideoExportJob, "run_id" | "status"> & {
+  output_dir: string;
+  status_url: string;
+};
+
+export async function videoDemoCapabilities() {
+  const { data } = await client.get("/capabilities", { timeout: 10000 });
+  return data;
+}
+
+export async function warmupVideoDemoModels(models: string[], modelVersions: Record<string, string | undefined> = {}) {
+  const { data } = await client.post("/warmup", { models, model_versions: modelVersions }, { timeout: 120000 });
+  return data;
+}
+
+export async function inferVideoFrame(form: FormData): Promise<FrameInferenceResponse> {
+  const { data } = await client.post("/infer-frame", form, { timeout: 45000 });
+  return data;
+}
+
+export async function createVideoExportJob(form: FormData): Promise<VideoExportCreateResponse> {
+  const { data } = await client.post("/export-jobs", form, { timeout: 10 * 60 * 1000 });
+  return data;
+}
+
+export async function videoExportJob(runId: string): Promise<VideoExportJob> {
+  const { data } = await client.get(`/export-jobs/${encodeURIComponent(runId)}`, { timeout: 10000 });
+  return data;
+}
+
+export function videoExportFileUrl(runId: string, fileName: string) {
+  return `${videoDemoBase}/api/v1/video-demo/export-jobs/${encodeURIComponent(runId)}/files/${encodeURIComponent(fileName)}`;
+}

frontend/src/views/alarms/AlarmsView.vue

@@ -17,14 +17,14 @@
       <el-table-column label="置信度" width="90"><template #default="scope">{{ Number(scope.row.confidence || 0).toFixed(2) }}</template></el-table-column>
       <el-table-column label="位置" min-width="150"><template #default="scope">{{ locationText(scope.row.location) }}</template></el-table-column>
       <el-table-column label="状态" width="105"><template #default="scope"><el-tag :type="statusTag(scope.row.status)" effect="plain">{{ statusLabel(scope.row.status) }}</el-tag></template></el-table-column>
-      <el-table-column label="工单" min-width="155"><template #default="scope"><span v-if="relatedWorkorder(scope.row.alarm_id)" class="entity-id">{{ relatedWorkorder(scope.row.alarm_id)?.workorder_id }}</span><span v-else>-</span></template></el-table-column>
+      <el-table-column label="工单" min-width="155"><template #default="scope"><el-button v-if="relatedWorkorder(scope.row.alarm_id)" link type="primary" class="entity-id" @click.stop="openWorkorder(scope.row)">{{ relatedWorkorder(scope.row.alarm_id)?.workorder_id }}</el-button><span v-else>-</span></template></el-table-column>
-      <el-table-column label="操作" width="230" fixed="right"><template #default="scope"><div class="list-actions" @click.stop><el-button link type="primary" @click="openDetail(scope.row)">证据链</el-button><el-button v-if="!scope.row.suppressed && !['confirmed','closed'].includes(scope.row.status)" link type="success" @click="decision(scope.row, 'confirm')">确认</el-button><el-button v-if="!scope.row.suppressed && scope.row.status !== 'closed'" link type="warning" @click="decision(scope.row, 'suppress')">抑制</el-button><el-button v-if="scope.row.suppressed" link type="primary" @click="decision(scope.row, 'reopen')">恢复</el-button></div></template></el-table-column>
+      <el-table-column label="操作" width="230" fixed="right"><template #default="scope"><div class="list-actions" @click.stop><el-button link type="primary" @click="openDetail(scope.row)">证据链</el-button><el-button v-if="canConfirm(scope.row)" link type="success" @click="decision(scope.row, 'confirm')">确认</el-button><el-button v-if="canSuppress(scope.row)" link type="warning" @click="decision(scope.row, 'suppress')">抑制</el-button><el-button v-if="scope.row.suppressed" link type="primary" @click="decision(scope.row, 'reopen')">恢复</el-button></div></template></el-table-column>
     </el-table><el-pagination v-model:current-page="page" v-model:page-size="pageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="[20, 50, 100]" :total="filteredRows.length" /></el-card>
 
     <el-drawer v-model="detailVisible" title="告警研判" size="min(760px, 94vw)">
       <el-skeleton v-if="detailLoading" :rows="10" animated />
       <AlarmEvidencePanel v-else-if="evidence.alarm" :evidence="evidence">
-        <template #actions><div class="drawer-actions"><el-button type="primary" @click="decision(evidence.alarm, 'confirm')">确认有效</el-button><el-button type="warning" @click="decision(evidence.alarm, 'suppress')">抑制告警</el-button><el-button @click="router.push({ path: '/gis', query: { alarmId: evidence.alarm.alarm_id } })">地图定位</el-button><el-button v-if="evidence.workorders?.length" @click="router.push(`/workorders/${evidence.workorders[0].workorder_id}`)">查看工单</el-button></div></template>
+        <template #actions><div class="drawer-actions"><el-button v-if="canConfirm(evidence.alarm)" type="primary" @click="decision(evidence.alarm, 'confirm')">确认有效</el-button><el-button v-if="canSuppress(evidence.alarm)" type="warning" @click="decision(evidence.alarm, 'suppress')">抑制告警</el-button><el-button v-if="evidence.alarm.suppressed" @click="decision(evidence.alarm, 'reopen')">恢复告警</el-button><el-button @click="openAlarmOnMap(evidence.alarm)">地图定位</el-button><el-button v-if="evidence.workorders?.length" @click="router.push(`/workorders/${evidence.workorders[0].workorder_id}`)">查看工单</el-button></div></template>
       </AlarmEvidencePanel>
       <el-empty v-else description="未获取到告警证据" />
     </el-drawer>
@@ -59,6 +59,10 @@ const filteredRows = computed(() => rows.value.filter((row) => matchTab(row) &&
 const pagedRows = computed(() => filteredRows.value.slice((page.value - 1) * pageSize.value, page.value * pageSize.value));
 const metrics = computed(() => [{ label: "有效告警", value: rows.value.filter((row) => !row.suppressed).length, note: "全部有效告警" }, { label: "严重与高等级", value: rows.value.filter((row) => ["critical", "high"].includes(String(row.severity)) && !row.suppressed).length, note: "优先处理" }, { label: "待研判", value: tabCount("pending"), note: "需要人工确认" }, { label: "已转工单", value: tabCount("dispatched"), note: "进入现场处置" }, { label: "已抑制", value: tabCount("suppressed"), note: "保留抑制原因" }]);
 function relatedWorkorder(id: unknown) { return workorderRows.value.find((row) => String(row.alarm_id) === String(id)); }
+function canConfirm(row: Row) { return !row.suppressed && !relatedWorkorder(row.alarm_id) && !["confirmed", "closed"].includes(String(row.status)); }
+function canSuppress(row: Row) { return !row.suppressed && !relatedWorkorder(row.alarm_id) && row.status !== "closed"; }
+function openWorkorder(row: Row) { const workorder = relatedWorkorder(row.alarm_id); if (workorder) router.push(`/workorders/${workorder.workorder_id}`); }
+function openAlarmOnMap(row: Row) { const workorder = relatedWorkorder(row.alarm_id); router.push({ path: "/gis", query: { alarmId: String(row.alarm_id), taskId: String(row.task_id || ""), workorderId: workorder ? String(workorder.workorder_id) : undefined } }); }
 function matchesTab(row: Row, tab: string) { if (tab === "all") return true; if (tab === "pending") return !row.suppressed && ["pending", "detected"].includes(String(row.status)); if (tab === "confirmed") return row.status === "confirmed"; if (tab === "suppressed") return Boolean(row.suppressed); if (tab === "dispatched") return Boolean(relatedWorkorder(row.alarm_id)); return row.status === "closed"; }
 function matchTab(row: Row) { return matchesTab(row, activeTab.value); }
 function tabCount(tab: string) { return rows.value.filter((row) => matchesTab(row, tab)).length; }

frontend/src/views/analysis/AnalysisView.vue

@@ -9,7 +9,7 @@
     <el-tabs v-model="activeTab" class="secondary-tabs">
       <el-tab-pane :label="`分析任务 ${jobRows.length}`" name="jobs">
         <div class="filter-bar"><el-input v-model="keyword" clearable placeholder="分析任务或巡检任务" :prefix-icon="Search" /><el-select v-model="statusFilter"><el-option label="全部状态" value="" /><el-option label="排队中" value="queued" /><el-option label="执行中" value="running" /><el-option label="已完成" value="completed" /><el-option label="失败" value="failed" /></el-select><span class="filter-spacer"></span><span>{{ filteredJobs.length }} 个分析任务</span></div>
-        <el-card class="workspace-card" shadow="never"><el-table :data="pagedJobs" empty-text="暂无分析任务"><el-table-column prop="analysis_job_id" label="分析任务" min-width="170"><template #default="scope"><span class="entity-id">{{ scope.row.analysis_job_id }}</span></template></el-table-column><el-table-column prop="task_id" label="巡检任务" min-width="170" show-overflow-tooltip /><el-table-column prop="line_id" label="线路" width="105" /><el-table-column label="资源" width="85"><template #default="scope">{{ arrayCount(scope.row.resource_ids) }}</template></el-table-column><el-table-column label="场景" width="85"><template #default="scope">{{ arrayCount(scope.row.scene_set) }}</template></el-table-column><el-table-column prop="analysis_mode" label="模式" width="90" /><el-table-column label="状态" width="105"><template #default="scope"><el-tag :type="statusTag(scope.row.status)" effect="plain">{{ statusLabel(scope.row.status) }}</el-tag></template></el-table-column><el-table-column label="操作" width="100"><template #default="scope"><el-button link type="primary" :disabled="scope.row.status === 'running'" @click="rerun(scope.row)">重新执行</el-button></template></el-table-column></el-table><el-pagination v-model:current-page="jobPage" v-model:page-size="jobPageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="filteredJobs.length" /></el-card>
+        <el-card class="workspace-card" shadow="never"><el-table :data="pagedJobs" empty-text="暂无分析任务"><el-table-column prop="analysis_job_id" label="分析任务" min-width="170"><template #default="scope"><span class="entity-id">{{ scope.row.analysis_job_id }}</span></template></el-table-column><el-table-column prop="task_id" label="巡检任务" min-width="170" show-overflow-tooltip /><el-table-column prop="line_id" label="线路" width="105" /><el-table-column label="资源" width="85"><template #default="scope">{{ arrayCount(scope.row.resource_ids) }}</template></el-table-column><el-table-column label="场景" width="85"><template #default="scope">{{ arrayCount(scope.row.scene_set) }}</template></el-table-column><el-table-column prop="analysis_mode" label="模式" width="90" /><el-table-column label="状态" width="105"><template #default="scope"><el-tag :type="statusTag(scope.row.status)" effect="plain">{{ statusLabel(scope.row.status) }}</el-tag></template></el-table-column><el-table-column label="操作" width="110"><template #default="scope"><el-button link type="primary" :loading="runningJobId === String(scope.row.analysis_job_id)" :disabled="!canRunJob(scope.row)" @click="rerun(scope.row)">{{ analysisActionLabel(scope.row) }}</el-button></template></el-table-column></el-table><el-pagination v-model:current-page="jobPage" v-model:page-size="jobPageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="filteredJobs.length" /></el-card>
       </el-tab-pane>
 
       <el-tab-pane :label="`AI 结果 ${resultRows.length}`" name="results">
@@ -23,7 +23,7 @@
 
       <el-tab-pane :label="`失败与降级 ${failedJobs.length}`" name="failed">
         <el-empty v-if="failedJobs.length === 0" description="当前没有失败或降级的分析任务" />
-        <el-card v-else class="workspace-card" shadow="never"><el-table :data="pagedFailedJobs"><el-table-column prop="analysis_job_id" label="分析任务" /><el-table-column prop="task_id" label="巡检任务" /><el-table-column prop="status" label="状态" /><el-table-column label="操作"><template #default="scope"><el-button link type="primary" @click="rerun(scope.row)">重新执行</el-button></template></el-table-column></el-table><el-pagination v-model:current-page="failedPage" v-model:page-size="failedPageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="failedJobs.length" /></el-card>
+        <el-card v-else class="workspace-card" shadow="never"><el-table :data="pagedFailedJobs"><el-table-column prop="analysis_job_id" label="分析任务" min-width="170" /><el-table-column prop="task_id" label="巡检任务" min-width="170" /><el-table-column prop="status" label="状态" width="100" /><el-table-column label="失败原因" min-width="220" show-overflow-tooltip><template #default="scope">{{ failureReason(scope.row.summary) }}</template></el-table-column><el-table-column label="操作" width="110"><template #default="scope"><el-button link type="primary" :loading="runningJobId === String(scope.row.analysis_job_id)" :disabled="!canRunJob(scope.row)" @click="rerun(scope.row)">{{ analysisActionLabel(scope.row) }}</el-button></template></el-table-column></el-table><el-pagination v-model:current-page="failedPage" v-model:page-size="failedPageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="failedJobs.length" /></el-card>
       </el-tab-pane>
     </el-tabs>
 
@@ -41,7 +41,7 @@ import { DEFAULT_PAGE_SIZES, usePagination } from "../../composables/usePaginati
 import { alarms, analysisJobs, analysisResults, resultEvidence, runAnalysisJob } from "../../services/api";
 import { safeObject, statusLabel, statusTag, type Row } from "../../types/demo-run";
 
-const route = useRoute(); const router = useRouter(); const loading = ref(false); const jobRows = ref<Row[]>([]); const resultRows = ref<Row[]>([]); const ruleRows = ref<Row[]>([]);
+const route = useRoute(); const router = useRouter(); const loading = ref(false); const jobRows = ref<Row[]>([]); const resultRows = ref<Row[]>([]); const ruleRows = ref<Row[]>([]); const runningJobId = ref("");
 const activeTab = ref(String(route.query.tab || (route.params.entityType === "results" ? "results" : "jobs"))); const keyword = ref(""); const statusFilter = ref(""); const sceneFilter = ref(""); const detailVisible = ref(false); const detailLoading = ref(false); const detail = ref<Row>({});
 const scenes = computed(() => [...new Set(resultRows.value.map((row) => String(row.scene)).filter(Boolean))]);
 const failedJobs = computed(() => jobRows.value.filter((row) => ["failed", "degraded"].includes(String(row.status))));
@@ -57,8 +57,11 @@ function parseArray(value: unknown): unknown[] { try { return Array.isArray(valu
 function arrayCount(value: unknown) { return parseArray(value).length; }
 function ruleText(value: unknown) { return parseArray(value).join(" / ") || "业务规则命中"; }
 function locationText(value: unknown) { const location = safeObject(value); return [location.mileage, location.distance_to_track_m ? `${location.distance_to_track_m}m` : ""].filter(Boolean).join(" / ") || "线路邻近"; }
+function failureReason(value: unknown) { return String(safeObject(value).error || "未记录失败原因"); }
+function canRunJob(row: Row) { return ["queued", "failed"].includes(String(row.status)) && runningJobId.value !== String(row.analysis_job_id); }
+function analysisActionLabel(row: Row) { const status = String(row.status); if (status === "queued") return "开始分析"; if (status === "failed") return "重新执行"; if (status === "running") return "执行中"; if (status === "completed") return "已完成"; return "不可执行"; }
 async function load() { loading.value = true; try { [jobRows.value, resultRows.value, ruleRows.value] = await Promise.all([analysisJobs(), analysisResults(), alarms(true)]); const id = String(route.params.entityId || ""); if (route.params.entityType === "results" && id) { const row = resultRows.value.find((item) => String(item.ai_result_id) === id); if (row) inspectResult(row); } } finally { loading.value = false; } }
-async function rerun(row: Row) { await runAnalysisJob(String(row.analysis_job_id)); ElMessage.success("分析任务已重新执行"); await load(); }
+async function rerun(row: Row) { if (!canRunJob(row)) return; const jobId = String(row.analysis_job_id); runningJobId.value = jobId; try { const result = await runAnalysisJob(jobId); ElMessage.success(result.status === "completed" ? "分析任务已完成" : "分析任务正在执行"); await load(); } catch { ElMessage.error("分析执行未成功，请查看任务状态与失败原因"); await load(); if (jobRows.value.find((item) => String(item.analysis_job_id) === jobId)?.status === "failed") activeTab.value = "failed"; } finally { runningJobId.value = ""; } }
 async function inspectResult(row: Row) { detailVisible.value = true; detailLoading.value = true; router.replace({ path: `/analysis/results/${row.ai_result_id}`, query: { ...route.query, tab: "results" } }); try { detail.value = await resultEvidence(String(row.ai_result_id)); } finally { detailLoading.value = false; } }
 watch(detailVisible, (visible) => { if (!visible && route.params.entityId) router.replace({ path: "/analysis", query: { ...route.query, tab: "results" } }); });
 watch([keyword, statusFilter, sceneFilter], () => { resetJobPage(); resetResultPage(); });

frontend/src/views/gis/GisView.vue

@@ -8,6 +8,8 @@
       <el-button type="primary" :disabled="!selectedTaskId" @click="exportThematicMap">导出专题图</el-button>
     </PageHeader>
 
+    <el-alert v-if="deepLinkLabel" class="deep-link-alert" type="info" :closable="false" show-icon title="深链定位已生效" :description="deepLinkLabel" />
+
     <div class="filter-bar">
       <el-select v-model="selectedTaskId" filterable placeholder="选择回放任务"><el-option v-for="task in taskRows" :key="task.task_id" :label="`${task.task_id} / ${task.line_id}`" :value="String(task.task_id)" /></el-select>
       <el-select v-model="exportFormat" class="export-format" placeholder="专题图格式"><el-option label="PDF 专题图" value="PDF" /><el-option label="PNG 图片" value="PNG" /><el-option label="HTML 可打印页" value="HTML" /><el-option label="GeoJSON 航迹" value="GEOJSON" /><el-option label="CSV 事件明细" value="CSV" /></el-select>
@@ -18,10 +20,10 @@
     </div>
 
     <section class="gis-page-layout">
-      <OperationalGisMap :feature-collection="filteredFeatureCollection" :layers="layers" :selected-id="String(selected?.alarm_id || '')" @select="selectFeature" />
+      <OperationalGisMap :feature-collection="filteredFeatureCollection" :layers="layers" :selected-id="selectedFeatureId" @select="selectFeature" />
       <aside class="gis-detail-panel">
         <template v-if="selected">
-          <div class="gis-detail-head"><div><el-tag :type="selected.severity === 'critical' ? 'danger' : 'warning'" effect="dark">{{ severityLabel(selected.severity) }}</el-tag><h2>{{ selected.scene }}</h2></div><el-button text :icon="Close" @click="selected = null" /></div>
+          <div class="gis-detail-head"><div><el-tag :type="selected.severity === 'critical' ? 'danger' : 'warning'" effect="dark">{{ severityLabel(selected.severity) }}</el-tag><h2>{{ selected.scene }}</h2></div><el-button text :icon="Close" @click="clearSelection" /></div>
           <el-descriptions :column="1" border size="small">
             <el-descriptions-item label="告警编号"><span class="entity-id">{{ selected.alarm_id }}</span></el-descriptions-item>
             <el-descriptions-item label="类别">{{ selected.category }}</el-descriptions-item>
@@ -32,6 +34,16 @@
           </el-descriptions>
           <div class="gis-detail-actions"><el-button type="primary" @click="router.push(`/alarms/${selected.alarm_id}`)">进入告警研判</el-button><el-button v-if="relatedWorkorder(selected.alarm_id)" @click="router.push(`/workorders/${relatedWorkorder(selected.alarm_id)?.workorder_id}`)">查看工单</el-button></div>
         </template>
+        <template v-else-if="selectedObject">
+          <div class="gis-detail-head"><div><el-tag type="success" effect="dark">巡检对象</el-tag><h2>{{ selectedObject.name }}</h2></div><el-button text :icon="Close" @click="clearSelection" /></div>
+          <el-descriptions :column="1" border size="small">
+            <el-descriptions-item label="对象编号"><span class="entity-id">{{ selectedObject.object_id }}</span></el-descriptions-item>
+            <el-descriptions-item label="对象类型">{{ selectedObject.object_type }}</el-descriptions-item>
+            <el-descriptions-item label="线路">{{ selectedObject.line_id || '-' }}</el-descriptions-item>
+            <el-descriptions-item label="风险等级">{{ selectedObject.risk_level || '-' }}</el-descriptions-item>
+            <el-descriptions-item label="状态">{{ selectedObject.status || '-' }}</el-descriptions-item>
+          </el-descriptions>
+        </template>
         <el-empty v-else description="点击地图告警点查看详情" :image-size="96" />
       </aside>
     </section>
@@ -71,7 +83,7 @@
 </template>
 
 <script setup lang="ts">
-import { computed, onBeforeUnmount, onMounted, reactive, ref } from "vue";
+import { computed, onBeforeUnmount, onMounted, reactive, ref, watch } from "vue";
 import { useRoute, useRouter } from "vue-router";
 import { ElMessage } from "element-plus";
 import { Close, Refresh } from "@element-plus/icons-vue";
@@ -86,7 +98,7 @@ const loading = ref(false);
 const alarmRows = ref<Row[]>([]); const taskRows = ref<Row[]>([]); const workorderRows = ref<Row[]>([]);
 const gisLayerRows = ref<Row[]>([]); const ruleSetRows = ref<Row[]>([]); const exemptionRows = ref<Row[]>([]); const geofenceRows = ref<Row[]>([]);
 const featureCollection = ref<Row>({ type: "FeatureCollection", features: [] });
-const selected = ref<Row | null>(null); const lineFilter = ref(""); const severityFilter = ref("");
+const selected = ref<Row | null>(null); const selectedObject = ref<Row | null>(null); const focusedTaskId = ref(""); const lineFilter = ref(""); const severityFilter = ref("");
 const spatialTab = ref("layers"); const spatialExplanation = ref<Row | null>(null);
 const selectedTaskId = ref(String(route.query.taskId || ""));
 const exportFormat = ref("PDF");
@@ -95,7 +107,15 @@ const playing = ref(false); const playbackSpeed = ref(1);
 let playbackTimer: ReturnType<typeof setInterval> | null = null;
 const layers = reactive({ routes: true, rules: true, alarms: true });
 const lines = computed(() => [...new Set(taskRows.value.map((row) => String(row.line_id)).filter(Boolean))]);
-const filteredAlarms = computed(() => alarmRows.value.filter((row) => (!severityFilter.value || row.severity === severityFilter.value) && (!lineFilter.value || taskLine(row.task_id) === lineFilter.value)));
+const filteredAlarms = computed(() => alarmRows.value.filter((row) => (!focusedTaskId.value || String(row.task_id) === focusedTaskId.value) && (!severityFilter.value || row.severity === severityFilter.value) && (!lineFilter.value || taskLine(row.task_id) === lineFilter.value)));
+const selectedFeatureId = computed(() => String(selected.value?.alarm_id || selectedObject.value?.object_id || ""));
+const deepLinkLabel = computed(() => {
+  if (route.query.workorderId) return `工单 ${route.query.workorderId} → 告警 ${selected.value?.alarm_id || "未找到空间点"}`;
+  if (route.query.alarmId) return `告警 ${route.query.alarmId}${focusedTaskId.value ? ` · 任务 ${focusedTaskId.value}` : ""}`;
+  if (route.query.objectId) return `巡检对象 ${route.query.objectId}`;
+  if (route.query.taskId) return `巡检任务 ${route.query.taskId}`;
+  return "";
+});
 const filteredFeatureCollection = computed(() => {
   const ids = new Set(filteredAlarms.value.map((row) => String(row.alarm_id)));
   const features: Row[] = (featureCollection.value.features || []).filter((feature: Row) => feature.properties?.domain_type !== "alarm" || ids.has(String(feature.properties?.alarm_id)));
@@ -143,7 +163,35 @@ function taskLine(taskId: unknown) { return String(taskRows.value.find((row) =>
 function relatedWorkorder(alarmId: unknown) { return workorderRows.value.find((row) => String(row.alarm_id) === String(alarmId)); }
 function locationText(row: Row) { const location = safeObject(row.location); return [location.mileage, location.distance_to_track_m ? `距线路 ${location.distance_to_track_m}m` : ""].filter(Boolean).join(" / ") || "线路邻近"; }
 function parseArray(value: unknown): string[] { try { return Array.isArray(value) ? value.map(String) : JSON.parse(String(value || "[]")); } catch { return []; } }
-function selectFeature(properties: Row) { selected.value = alarmRows.value.find((row) => String(row.alarm_id) === String(properties.alarm_id)) || properties; }
+function selectFeature(properties: Row) {
+  if (properties.domain_type === "inspection_object") {
+    selected.value = null; selectedObject.value = properties; focusedTaskId.value = "";
+    router.replace({ path: "/gis", query: { objectId: String(properties.object_id) } });
+    return;
+  }
+  const alarm = alarmRows.value.find((row) => String(row.alarm_id) === String(properties.alarm_id)) || properties;
+  selected.value = alarm; selectedObject.value = null; focusedTaskId.value = String(alarm.task_id || ""); selectedTaskId.value = focusedTaskId.value || selectedTaskId.value;
+  const workorder = relatedWorkorder(alarm.alarm_id);
+  router.replace({ path: "/gis", query: { alarmId: String(alarm.alarm_id), taskId: focusedTaskId.value || undefined, workorderId: workorder ? String(workorder.workorder_id) : undefined } });
+}
+function clearSelection() { selected.value = null; selectedObject.value = null; focusedTaskId.value = ""; router.replace({ path: "/gis" }); }
+function applyDeepLink() {
+  const workorderId = String(route.query.workorderId || "");
+  const linkedWorkorder = workorderRows.value.find((row) => String(row.workorder_id) === workorderId);
+  const alarmId = String(route.query.alarmId || linkedWorkorder?.alarm_id || "");
+  const alarm = alarmRows.value.find((row) => String(row.alarm_id) === alarmId);
+  const taskId = String(route.query.taskId || linkedWorkorder?.task_id || alarm?.task_id || "");
+  focusedTaskId.value = taskId;
+  if (taskId) selectedTaskId.value = taskId;
+  if (alarm) { selected.value = alarm; selectedObject.value = null; return; }
+  const objectId = String(route.query.objectId || "");
+  if (objectId) {
+    const objectFeature = (featureCollection.value.features || []).find((feature: Row) => String(feature.id || feature.properties?.object_id) === objectId && feature.properties?.domain_type === "inspection_object");
+    selected.value = null; selectedObject.value = objectFeature ? { ...objectFeature.properties, object_id: objectId } : { object_id: objectId, name: "未找到巡检对象" }; return;
+  }
+  selectedObject.value = null;
+  selected.value = taskId ? alarmRows.value.find((row) => String(row.task_id) === taskId) || null : filteredAlarms.value[0] || null;
+}
 function formatDate(value: unknown) { return value ? new Date(String(value)).toLocaleString("zh-CN", { hour12: false }) : "-"; }
 async function loadPlayback() {
   if (!selectedTaskId.value) return;
@@ -199,7 +247,8 @@ async function explainSelected() {
   spatialTab.value = "rules";
   ElMessage.success(`空间解释完成：${result.decision}`);
 }
-async function load() { loading.value = true; try { [alarmRows.value, taskRows.value, workorderRows.value, featureCollection.value, gisLayerRows.value, ruleSetRows.value, exemptionRows.value, geofenceRows.value] = await Promise.all([alarms(), tasks(), workorders(), gisFeatures(), gisLayers(), spatialRuleSets(), spatialExemptions(), geofences()]); if (!selectedTaskId.value) selectedTaskId.value = String(taskRows.value[0]?.task_id || ""); selected.value = filteredAlarms.value[0] || null; } finally { loading.value = false; } }
+async function load() { loading.value = true; try { [alarmRows.value, taskRows.value, workorderRows.value, featureCollection.value, gisLayerRows.value, ruleSetRows.value, exemptionRows.value, geofenceRows.value] = await Promise.all([alarms(true), tasks(), workorders(), gisFeatures(), gisLayers(), spatialRuleSets(), spatialExemptions(), geofences()]); if (!selectedTaskId.value) selectedTaskId.value = String(taskRows.value[0]?.task_id || ""); applyDeepLink(); } finally { loading.value = false; } }
+watch(() => route.query, applyDeepLink);
 onMounted(load);
 onBeforeUnmount(pausePlayback);
 </script>
@@ -217,4 +266,5 @@ onBeforeUnmount(pausePlayback);
 .export-format { width: 150px; }
 .spatial-card { margin-top: 18px; }
 .spatial-explanation { margin-top: 14px; }
+.deep-link-alert { margin-bottom: 16px; }
 </style>

frontend/src/views/operations/OperationsView.vue

@@ -14,13 +14,19 @@
     <el-tabs v-model="activeTab" class="secondary-tabs">
       <el-tab-pane label="服务状态" name="services">
         <section class="service-grid">
-          <div v-for="service in services" :key="service.name" class="service-card">
+          <div v-for="service in services" :key="String(service.code)" class="service-card">
             <header>
-              <el-icon><component :is="service.icon" /></el-icon>
+              <el-icon><component :is="serviceIcon(service.code)" /></el-icon>
               <span><strong>{{ service.name }}</strong><small>{{ service.endpoint }}</small></span>
-              <i class="status-dot" :class="{ danger: health.status !== 'UP' }"></i>
+              <i class="status-dot" :class="serviceStatusClass(service.status)"></i>
             </header>
-            <dl><div><dt>状态</dt><dd>{{ health.status === 'UP' ? '运行正常' : '连接异常' }}</dd></div><div><dt>检查方式</dt><dd>{{ service.check }}</dd></div></dl>
+            <dl>
+              <div><dt>状态</dt><dd>{{ service.status_text || statusText(service.status) }}</dd></div>
+              <div><dt>响应</dt><dd>{{ latencyText(service.latency_ms) }}</dd></div>
+              <div><dt>检查方式</dt><dd>{{ service.check }}</dd></div>
+              <div><dt>检查时间</dt><dd>{{ timeOnly(service.checked_at) }}</dd></div>
+            </dl>
+            <p class="service-detail">{{ service.detail || "暂无探测详情" }}</p>
           </div>
         </section>
 
@@ -125,7 +131,18 @@
       </el-tab-pane>
 
       <el-tab-pane label="接口集成" name="integrations">
-        <el-card class="workspace-card" shadow="never"><el-table :data="integrations"><el-table-column prop="name" label="集成系统" min-width="150" /><el-table-column prop="type" label="方式" width="110" /><el-table-column prop="endpoint" label="地址或通道" min-width="220" /><el-table-column label="状态" width="110"><template #default><el-tag type="success" effect="plain">已连接</el-tag></template></el-table-column><el-table-column prop="description" label="用途" min-width="200" /></el-table></el-card>
+        <el-card class="workspace-card" shadow="never">
+          <el-table :data="integrations" empty-text="尚未获得集成探测结果">
+            <el-table-column prop="name" label="集成系统" min-width="150" />
+            <el-table-column prop="type" label="方式" width="110" />
+            <el-table-column prop="endpoint" label="地址或通道" min-width="220" />
+            <el-table-column label="状态" width="120">
+              <template #default="scope"><el-tag :type="integrationTagType(scope.row.status)" effect="plain">{{ scope.row.status_text || "未探测" }}</el-tag></template>
+            </el-table-column>
+            <el-table-column prop="description" label="用途" min-width="190" />
+            <el-table-column prop="evidence" label="状态依据" min-width="260" />
+          </el-table>
+        </el-card>
       </el-tab-pane>
       <el-tab-pane label="组织与责任" name="organization">
         <div class="page-grid two-columns">
@@ -201,11 +218,11 @@ import {
   modelArtifactJob,
   modelArtifactRuntimeOverview,
   modelArtifactUninstallImpact,
+  operationsHealth,
   organizations as loadOrganizations,
   overview,
   platformEvents,
   responsibilityRules,
-  serviceHealth,
   uninstallModelArtifact,
   unloadModelArtifact,
   uploadModelArtifact
@@ -215,7 +232,7 @@ import { safeObject, type Row } from "../../types/demo-run";
 const loading = ref(false);
 const activeTab = ref("services");
 const runtimeScope = ref<"active" | "all">("active");
-const health = ref<Row>({ status: "UNKNOWN" });
+const operations = ref<Row>({ status: "UNKNOWN", services: [], integrations: [] });
 const stats = ref<Row>({});
 const eventRows = ref<Row[]>([]);
 const organizationRows = ref<Row[]>([]);
@@ -243,12 +260,14 @@ const { currentPage: auditPage, pageSize: auditPageSize, pagedItems: pagedAuditR
 const installerInfo = computed(() => safeObject(artifactOverview.value.installer));
 const installerStatus = computed(() => String(installerInfo.value.status || "unavailable"));
 const installerMessage = computed(() => String(installerInfo.value.message || "模型制品安装代理当前不可用，请检查 8103 服务。"));
-const artifactStoreAvailable = computed(() => Boolean(artifactOverview.value.artifact_store_available));
+const services = computed<Row[]>(() => asRows(operations.value.services));
+const integrations = computed<Row[]>(() => asRows(operations.value.integrations));
+const artifactStoreAvailable = computed(() => services.value.some(service => service.code === "minio" && service.status === "UP"));
 const readyCount = computed(() => artifactRows.value.filter(row => row.package_format === "builtin" || row.installation_status === "ready").length);
 const runtimeProfile = computed(() => String(installerInfo.value.profile || visionRuntime.value.profile || spatialRuntime.value.profile || "未配置"));
 const executionProvider = computed(() => String(visionRuntime.value.execution_provider || spatialRuntime.value.execution_provider || "-"));
 const metrics = computed(() => [
-  { label: "平台状态", value: health.value.status === "UP" ? "正常" : "异常", note: "Actuator 健康检查" },
+  { label: "服务状态", value: overallStatusText.value, note: checkedAtText.value },
   { label: "模型制品", value: `${readyCount.value}/${artifactRows.value.length}`, note: runtimeProfile.value },
   { label: "业务对象", value: Number(stats.value.tasks || 0) + Number(stats.value.resources || 0), note: "任务与资源" },
   { label: "分析结果", value: stats.value.ai_results || 0, note: "累计推理结果" },
@@ -262,9 +281,19 @@ const jobStatusType = computed(() => currentJob.value.status === "failed" ? "dan
 const jobStatusText = computed(() => ({ queued: "排队中", running: "执行中", completed: "已完成", failed: "失败" }[String(currentJob.value.status)] || "等待状态"));
 const jobTitle = computed(() => ({ install: "模型制品安装", load: "模型加载", unload: "运行资源释放", uninstall: "模型制品卸载" }[String(currentJob.value.job_type)] || "模型制品任务"));
 const latestJobMessage = computed(() => String(jobEvents.value[jobEvents.value.length - 1]?.message || "正在等待任务执行"));
-
+const overallStatusText = computed(() => ({ UP: "全部正常", DEGRADED: "部分异常", DOWN: "不可用" }[String(operations.value.status)] || "未检查"));
-const services = [{ name: "业务平台", endpoint: ":8080", check: "Actuator", icon: SetUp }, { name: "视觉推理服务", endpoint: ":8101", check: "HTTP Health", icon: Cpu }, { name: "点云分析服务", endpoint: ":8102", check: "HTTP Health", icon: DataLine }, { name: "模型安装代理", endpoint: ":8103", check: "HTTP Health", icon: Download }, { name: "PostGIS / Kafka / Redis / MinIO", endpoint: "基础设施", check: "容器健康检查", icon: Connection }];
+const checkedAtText = computed(() => operations.value.checked_at ? `检查于 ${formatDateTime(operations.value.checked_at)}` : "等待首次检查");
-const integrations = [{ name: "无人机任务接口", type: "SDK/API", endpoint: "/api/v1/uav/callbacks/mission-status", description: "航线任务和飞行状态回调" }, { name: "多源数据接入", type: "REST", endpoint: "/api/v1/inspection/resources/complete", description: "资源清单与接入完成通知" }, { name: "工单系统", type: "REST", endpoint: "/api/v1/workorders/callbacks/status", description: "工单处置状态同步" }, { name: "业务事件总线", type: "Kafka", endpoint: "platform-events", description: "任务、告警、工单和样本事件" }];
+const serviceIconMap: Record<string, typeof SetUp> = {
+  platform: SetUp,
+  postgis: DataLine,
+  redis: Connection,
+  kafka: Connection,
+  minio: Download,
+  vision: Cpu,
+  pointcloud: DataLine,
+  "artifact-installer": Download,
+  "uav-access": Connection
+};
 const configGroups = [{ title: "数据基础设施", description: "持久化、缓存、消息和对象存储", items: [{ label: "数据库", value: "PostgreSQL 16 + PostGIS 3.4" }, { label: "缓存", value: "Redis 7" }, { label: "消息", value: "Kafka 3.8.1" }, { label: "对象存储", value: "MinIO S3" }] }, { title: "算法服务", description: "视觉、空间分析与制品管理", items: [{ label: "视觉推理", value: "FastAPI :8101" }, { label: "点云分析", value: "FastAPI :8102" }, { label: "制品安装", value: "FastAPI :8103" }, { label: "模型交换", value: "ONNX Runtime / TensorRT 适配" }] }, { title: "平台与监控", description: "业务服务和可观测性", items: [{ label: "业务平台", value: "Spring Boot 3.3" }, { label: "前端", value: "Vue 3 + TypeScript" }, { label: "指标", value: "Prometheus :9090" }] }];
 
 function openUpload(row: Row) {
@@ -336,13 +365,41 @@ function environmentText(value: unknown) { return ({ local: "本地", developmen
 function payloadText(value: unknown) { const data = safeObject(value); return Object.entries(data).slice(0, 4).map(([key, item]) => `${key}: ${item}`).join("，") || "-"; }
 function asRows(value: unknown): Row[] { return Array.isArray(value) ? value.map(item => safeObject(item)) : []; }
 function formatDateTime(value: unknown) { return value ? new Date(String(value)).toLocaleString("zh-CN", { hour12: false }) : "-"; }
+function timeOnly(value: unknown) { return value ? new Date(String(value)).toLocaleTimeString("zh-CN", { hour12: false }) : "-"; }
+function latencyText(value: unknown) { const latency = Number(value); return Number.isFinite(latency) ? `${latency} ms` : "-"; }
+function statusText(value: unknown) { return value === "UP" ? "运行正常" : value === "DOWN" ? "连接异常" : "未探测"; }
+function serviceStatusClass(value: unknown) { return { danger: value === "DOWN", warning: value !== "UP" && value !== "DOWN" }; }
+function integrationTagType(value: unknown) { return value === "AVAILABLE" ? "success" : value === "UNAVAILABLE" ? "danger" : "info"; }
+function serviceIcon(value: unknown) { return serviceIconMap[String(value)] || Connection; }
 function errorMessage(error: unknown) { const item = error as { response?: { data?: { message?: string; detail?: string; error?: string } }; message?: string }; return item.response?.data?.message || item.response?.data?.detail || item.response?.data?.error || item.message || "操作失败"; }
-function openPrometheus() { window.open("http://localhost:9090", "_blank", "noopener,noreferrer"); }
+function openPrometheus() { const target = new URL(window.location.href); target.port = "9090"; target.pathname = "/"; target.search = ""; target.hash = ""; window.open(target.toString(), "_blank", "noopener,noreferrer"); }
 async function load() {
   loading.value = true;
   try {
-    const [healthData, statData, platformEventData, runtimeData, artifactData, organizationData, responsibilityData] = await Promise.all([serviceHealth(), overview(), platformEvents(), aiRuntimeStatus(), modelArtifactRuntimeOverview(), loadOrganizations(), responsibilityRules()]);
+    const loaders: Array<{ name: string; request: Promise<unknown>; apply: (value: unknown) => void }> = [
-    health.value = healthData; stats.value = statData; eventRows.value = platformEventData; runtime.value = runtimeData; artifactOverview.value = artifactData; organizationRows.value = organizationData; responsibilityRows.value = responsibilityData;
+      { name: "服务健康", request: operationsHealth(), apply: value => operations.value = safeObject(value) },
+      { name: "业务统计", request: overview(), apply: value => stats.value = safeObject(value) },
+      { name: "平台事件", request: platformEvents(), apply: value => eventRows.value = asRows(value) },
+      { name: "模型运行时", request: aiRuntimeStatus(), apply: value => runtime.value = safeObject(value) },
+      { name: "模型制品", request: modelArtifactRuntimeOverview(), apply: value => artifactOverview.value = safeObject(value) },
+      { name: "组织结构", request: loadOrganizations(), apply: value => organizationRows.value = asRows(value) },
+      { name: "责任规则", request: responsibilityRules(), apply: value => responsibilityRows.value = asRows(value) }
+    ];
+    const results = await Promise.allSettled(loaders.map(item => item.request));
+    const failures: string[] = [];
+    results.forEach((result, index) => {
+      if (result.status === "fulfilled") loaders[index].apply(result.value);
+      else failures.push(loaders[index].name);
+    });
+    if (results[0].status === "rejected") {
+      operations.value = {
+        status: "DOWN",
+        checked_at: new Date().toISOString(),
+        services: [{ code: "platform", name: "业务平台", endpoint: "/api/v1/operations/health", check: "HTTP", status: "DOWN", status_text: "连接异常", detail: errorMessage(results[0].reason), checked_at: new Date().toISOString() }],
+        integrations: []
+      };
+    }
+    if (failures.length) ElMessage.warning(`刷新完成，但以下数据不可用：${failures.join("、")}`);
   } catch (error) { ElMessage.error(errorMessage(error)); }
   finally { loading.value = false; }
 }
@@ -353,6 +410,7 @@ onBeforeUnmount(() => window.clearTimeout(jobTimer));
 
 <style scoped>
 .runtime-panel { margin-top: 18px; border-top: 1px solid var(--el-border-color-light); padding-top: 16px; }
+.service-detail { margin: 10px 0 0; min-height: 34px; color: var(--el-text-color-secondary); font-size: 12px; line-height: 1.5; word-break: break-word; }
 .runtime-panel > header, .runtime-panel > footer { display: flex; align-items: center; justify-content: space-between; gap: 16px; }
 .runtime-panel > header { margin-bottom: 12px; }
 .runtime-panel > header > div:first-child, .model-cell, .dialog-context { display: grid; gap: 3px; }

frontend/src/views/planning/PlanningView.vue

@@ -84,7 +84,7 @@
       <template #footer><el-button @click="objectDialog = false">取消</el-button><el-button type="primary" :loading="saving" @click="saveObject">创建</el-button></template>
     </el-dialog>
 
-    <el-drawer v-model="detailVisible" title="计划详情" size="720px"><template v-if="selectedPlan"><el-descriptions :column="1" border><el-descriptions-item label="计划编号"><span class="entity-id">{{ selectedPlan.plan_id }}</span></el-descriptions-item><el-descriptions-item label="巡检方式">{{ triggerTypeLabel(selectedPlan.trigger_type) }}</el-descriptions-item><el-descriptions-item label="业务类型">{{ planTypeLabel(selectedPlan.plan_type) }}</el-descriptions-item><el-descriptions-item label="执行安排">{{ scheduleText(selectedPlan) }}</el-descriptions-item><el-descriptions-item label="下次/计划执行">{{ executionDateText(selectedPlan) }}</el-descriptions-item><el-descriptions-item label="优先级">{{ priorityLabel(selectedPlan.priority) }}</el-descriptions-item><el-descriptions-item label="对象">{{ objectNames(selectedPlan.object_ids) }}</el-descriptions-item><el-descriptions-item label="场景">{{ parseArray(selectedPlan.scene_set).join('、') }}</el-descriptions-item><el-descriptions-item label="已生成任务">{{ selectedPlan.generated_task_count }}</el-descriptions-item></el-descriptions><div class="subsection-head drawer-section"><div><strong>执行预览</strong><span>使用当前调度配置计算后续窗口</span></div><el-button link @click="loadPlanRuntime(selectedPlan)">刷新</el-button></div><el-table :data="previewRunsRows" size="small" empty-text="暂无预览"><el-table-column prop="sequence" label="#" width="52" /><el-table-column label="窗口开始" min-width="160"><template #default="scope">{{ formatDate(scope.row.scheduled_window_start) }}</template></el-table-column><el-table-column label="窗口结束" min-width="160"><template #default="scope">{{ formatDate(scope.row.scheduled_window_end) }}</template></el-table-column><el-table-column label="生成键" min-width="170"><template #default="scope"><span class="entity-id">{{ String(scope.row.generation_key).slice(0, 18) }}</span></template></el-table-column></el-table><div class="subsection-head drawer-section"><div><strong>执行记录</strong><span>后台调度和手动触发均留痕</span></div></div><el-table :data="executionRows" size="small" empty-text="暂无执行记录"><el-table-column prop="status" label="状态" width="110" /><el-table-column label="窗口" min-width="185"><template #default="scope">{{ formatDate(scope.row.scheduled_window_start) }}</template></el-table-column><el-table-column label="任务" min-width="120"><template #default="scope">{{ parseArray(scope.row.generated_task_ids).length }} 个</template></el-table-column><el-table-column prop="trigger_source" label="来源" width="110" /></el-table></template></el-drawer>
+    <el-drawer v-model="detailVisible" title="计划详情" size="720px"><template v-if="selectedPlan"><el-descriptions :column="1" border><el-descriptions-item label="计划编号"><span class="entity-id">{{ selectedPlan.plan_id }}</span></el-descriptions-item><el-descriptions-item label="巡检方式">{{ triggerTypeLabel(selectedPlan.trigger_type) }}</el-descriptions-item><el-descriptions-item label="业务类型">{{ planTypeLabel(selectedPlan.plan_type) }}</el-descriptions-item><el-descriptions-item label="执行安排">{{ scheduleText(selectedPlan) }}</el-descriptions-item><el-descriptions-item label="下次/计划执行">{{ executionDateText(selectedPlan) }}</el-descriptions-item><el-descriptions-item label="优先级">{{ priorityLabel(selectedPlan.priority) }}</el-descriptions-item><el-descriptions-item label="对象">{{ objectNames(selectedPlan.object_ids) }}</el-descriptions-item><el-descriptions-item label="场景">{{ parseArray(selectedPlan.scene_set).join('、') }}</el-descriptions-item><el-descriptions-item label="已生成任务">{{ selectedPlan.generated_task_count }}</el-descriptions-item></el-descriptions><div class="subsection-head drawer-section"><div><strong>执行预览</strong><span>{{ selectedPlan.trigger_type === 'MANUAL' ? '手动巡检仅预览本次执行窗口' : '使用当前调度配置计算后续窗口' }}</span></div><el-button link @click="loadPlanRuntime(selectedPlan)">刷新</el-button></div><el-table :data="previewRunsRows" size="small" empty-text="暂无预览"><el-table-column prop="sequence" label="#" width="52" /><el-table-column label="窗口开始" min-width="160"><template #default="scope">{{ formatDate(scope.row.scheduled_window_start) }}</template></el-table-column><el-table-column label="窗口结束" min-width="160"><template #default="scope">{{ formatDate(scope.row.scheduled_window_end) }}</template></el-table-column><el-table-column label="生成键" min-width="170"><template #default="scope"><span class="entity-id">{{ String(scope.row.generation_key).slice(0, 18) }}</span></template></el-table-column></el-table><div class="subsection-head drawer-section"><div><strong>执行记录</strong><span>后台调度和手动触发均留痕</span></div></div><el-table :data="executionRows" size="small" empty-text="暂无执行记录"><el-table-column prop="status" label="状态" width="110" /><el-table-column label="窗口" min-width="185"><template #default="scope">{{ formatDate(scope.row.scheduled_window_start) }}</template></el-table-column><el-table-column label="任务" min-width="120"><template #default="scope">{{ parseArray(scope.row.generated_task_ids).length }} 个</template></el-table-column><el-table-column prop="trigger_source" label="来源" width="110" /></el-table></template></el-drawer>
 
     <el-drawer v-model="versionVisible" title="对象版本" size="620px">
       <template v-if="selectedObject"><h3>{{ selectedObject.name }}</h3><p class="entity-id">{{ selectedObject.object_id }}</p><el-table :data="objectVersionRows" size="small" empty-text="暂无版本记录"><el-table-column label="版本" width="70"><template #default="scope">V{{ scope.row.version_no }}</template></el-table-column><el-table-column prop="change_type" label="变更" width="95" /><el-table-column prop="change_reason" label="原因" min-width="140" /><el-table-column label="来源" min-width="150"><template #default="scope"><span class="entity-id">{{ scope.row.source_job_id || '-' }}</span></template></el-table-column><el-table-column label="时间" min-width="155"><template #default="scope">{{ formatDate(scope.row.created_at) }}</template></el-table-column></el-table></template>

frontend/src/views/resources/ResourcesView.vue

@@ -35,7 +35,7 @@
           <div v-for="processor in processors" :key="processor.type"><span>{{ processor.type }}</span><strong>{{ typeCount(processor.key) }}</strong><small>{{ processor.steps }}</small></div>
         </div>
         <el-card class="workspace-card" shadow="never"><el-table :data="pagedPreprocessRows" empty-text="暂无预处理记录">
-          <el-table-column prop="resource_id" label="资源" min-width="170"><template #default="scope"><span class="entity-id">{{ scope.row.resource_id }}</span></template></el-table-column><el-table-column label="处理链" min-width="280"><template #default="scope">{{ processorFor(scope.row.resource_type) }}</template></el-table-column><el-table-column prop="pipeline_version" label="流水线版本" min-width="145" /><el-table-column label="状态" width="110"><template #default="scope"><el-tag :type="scope.row.status === 'COMPLETED' ? 'success' : 'warning'" effect="plain">{{ scope.row.status === 'COMPLETED' ? '已完成' : '待处理' }}</el-tag></template></el-table-column><el-table-column label="质量" width="100"><template #default="scope">{{ scope.row.quality_status }}</template></el-table-column><el-table-column label="操作" width="120"><template #default="scope"><el-button link type="primary" :disabled="scope.row.status === 'COMPLETED'" @click="runPreprocess(scope.row)">执行预处理</el-button></template></el-table-column>
+          <el-table-column prop="resource_id" label="资源" min-width="170"><template #default="scope"><span class="entity-id">{{ scope.row.resource_id }}</span></template></el-table-column><el-table-column label="处理链" min-width="280"><template #default="scope">{{ processorFor(scope.row.resource_type) }}</template></el-table-column><el-table-column prop="pipeline_version" label="流水线版本" min-width="145" /><el-table-column label="状态" width="110"><template #default="scope"><el-tag :type="scope.row.status === 'COMPLETED' ? 'success' : 'warning'" effect="plain">{{ scope.row.status === 'COMPLETED' ? '已完成' : '待处理' }}</el-tag></template></el-table-column><el-table-column label="质量" width="100"><template #default="scope">{{ scope.row.quality_status }}</template></el-table-column><el-table-column label="操作" width="220"><template #default="scope"><div class="list-actions"><el-button link type="primary" :disabled="scope.row.status === 'COMPLETED'" @click="runPreprocess(scope.row)">执行预处理</el-button><el-button link type="primary" :loading="Boolean(analysisRunning[String(scope.row.resource_id)])" :disabled="!canStartAnalysis(scope.row)" @click="startAnalysis(scope.row)">{{ analysisActionLabel(scope.row) }}</el-button></div></template></el-table-column>
         </el-table><el-pagination v-model:current-page="preprocessPage" v-model:page-size="preprocessPageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="preprocessRows.length" /></el-card>
       </el-tab-pane>
 
@@ -65,12 +65,13 @@ import { ElMessage } from "element-plus";
 import { Grid, Histogram, Picture, Refresh, Search, Upload, UploadFilled } from "@element-plus/icons-vue";
 import PageHeader from "../../components/common/PageHeader.vue";
 import { DEFAULT_PAGE_SIZES, usePagination } from "../../composables/usePagination";
-import { preprocessJobs, resourcePreview, resources, runPreprocessJob, tasks, uploadInspectionFile } from "../../services/api";
+import { createAnalysisJob, preprocessJobs, resourcePreview, resources, runAnalysisJob, runPreprocessJob, tasks, uploadInspectionFile } from "../../services/api";
 import { safeObject, statusLabel, statusTag, type Row } from "../../types/demo-run";
 
 const route = useRoute(); const router = useRouter(); const loading = ref(false); const rows = ref<Row[]>([]); const preprocessRows = ref<Row[]>([]); const taskRows = ref<Row[]>([]); const activeTab = ref(String(route.query.tab || "library")); const keyword = ref(""); const typeFilter = ref("");
 const detailVisible = ref(false); const detailLoading = ref(false); const detail = ref<Row>({});
 const uploadVisible = ref(false); const uploading = ref(false); const uploadProgress = ref(0); const uploadFile = ref<File | null>(null); const uploadForm = reactive({ task_id: "", resource_type: "image" });
+const analysisRunning = ref<Record<string, boolean>>({});
 const resourceTypes = [{ value: "image", label: "可见光" }, { value: "thermal", label: "红外" }, { value: "pointcloud", label: "点云" }, { value: "tif", label: "TIF/DEM" }];
 const processors = [{ key: "image", type: "可见光", steps: "畸变校正 / 质量检查 / 标准化" }, { key: "thermal", type: "红外", steps: "温度标定 / 坏点修复 / 辐射校正" }, { key: "pointcloud", type: "点云", steps: "去噪 / 配准 / 分割 / 坐标转换" }, { key: "tif", type: "TIF/DEM", steps: "投影校验 / 重采样 / 高程转换" }];
 const filteredRows = computed(() => rows.value.filter((row) => (!typeFilter.value || row.resource_type === typeFilter.value) && (!route.query.taskId || row.task_id === route.query.taskId) && (!keyword.value || `${row.resource_id} ${row.task_id} ${row.storage_url}`.toLowerCase().includes(keyword.value.toLowerCase()))));
@@ -86,11 +87,15 @@ function fileName(path: unknown) { return String(path || "-").split("/").pop() |
 function metadataSummary(value: unknown) { const data = safeObject(value); return Object.entries(data).slice(0, 2).map(([key, item]) => `${key}: ${item}`).join("，") || "无扩展元数据"; }
 function processorFor(type: unknown) { return processors.find((item) => item.key === String(type))?.steps || "格式校验 / 标准化"; }
 function formatDateTime(value: unknown) { return value ? new Date(String(value)).toLocaleString("zh-CN", { hour12: false }) : "-"; }
+function resourceAnalysisStatus(row: Row) { return String(rows.value.find((item) => item.resource_id === row.resource_id)?.analysis_status || "pending"); }
+function canStartAnalysis(row: Row) { return row.status === "COMPLETED" && row.quality_status === "PASSED" && resourceAnalysisStatus(row) !== "completed" && !analysisRunning.value[String(row.resource_id)]; }
+function analysisActionLabel(row: Row) { if (resourceAnalysisStatus(row) === "completed") return "已分析"; if (row.status !== "COMPLETED") return "等待预处理"; if (row.quality_status !== "PASSED") return "质量未通过"; return "创建并分析"; }
 async function load() { loading.value = true; try { [rows.value, preprocessRows.value, taskRows.value] = await Promise.all([resources(), preprocessJobs(), tasks()]); if (!uploadForm.task_id && taskRows.value.length) uploadForm.task_id = String(taskRows.value[0].task_id); const id = String(route.params.resourceId || ""); if (id) { const row = rows.value.find((item) => String(item.resource_id) === id); if (row) inspect(row); } } finally { loading.value = false; } }
 function selectFile(file: UploadFile) { uploadFile.value = file.raw || null; }
 function removeFile() { uploadFile.value = null; }
-async function uploadResource() { if (!uploadForm.task_id || !uploadFile.value) return ElMessage.warning("请选择巡检任务和资源文件"); uploading.value = true; uploadProgress.value = 0; try { const result = await uploadInspectionFile(uploadForm.task_id, uploadFile.value, uploadForm.resource_type, (value) => uploadProgress.value = value); await runPreprocessJob(String(result.preprocess_job_id)); ElMessage.success("资源已分片上传、摘要校验并完成基础预处理"); uploadVisible.value = false; uploadFile.value = null; activeTab.value = "preprocess"; await load(); } finally { uploading.value = false; } }
+async function uploadResource() { if (!uploadForm.task_id || !uploadFile.value) return ElMessage.warning("请选择巡检任务和资源文件"); uploading.value = true; uploadProgress.value = 0; try { const result = await uploadInspectionFile(uploadForm.task_id, uploadFile.value, uploadForm.resource_type, (value) => uploadProgress.value = value); await runPreprocessJob(String(result.preprocess_job_id)); ElMessage.success("资源已完成上传与预处理，可在预处理记录中创建 AI 分析"); uploadVisible.value = false; uploadFile.value = null; activeTab.value = "preprocess"; await load(); } finally { uploading.value = false; } }
-async function runPreprocess(row: Row) { await runPreprocessJob(String(row.preprocess_job_id)); ElMessage.success("预处理与质量检查已完成"); await load(); }
+async function runPreprocess(row: Row) { await runPreprocessJob(String(row.preprocess_job_id)); ElMessage.success("预处理与质量检查已完成，可继续创建 AI 分析"); await load(); }
+async function startAnalysis(row: Row) { if (!canStartAnalysis(row)) return; const resourceId = String(row.resource_id); let jobId = ""; analysisRunning.value[resourceId] = true; try { const job = await createAnalysisJob({ task_id: String(row.task_id), resource_ids: [resourceId], analysis_mode: "offline", priority: "normal" }); jobId = String(job.analysis_job_id); await runAnalysisJob(jobId); ElMessage.success("AI 分析已完成，正在打开结果"); await router.push({ path: "/analysis", query: { tab: "results", resourceId } }); } catch { if (jobId) { ElMessage.error("AI 分析失败，任务状态已记录，可在智能分析中重试"); await router.push({ path: "/analysis", query: { tab: "failed" } }); } else { ElMessage.error("分析任务创建失败，请稍后重试"); } } finally { analysisRunning.value[resourceId] = false; } }
 async function inspect(row: Row) { detailVisible.value = true; detailLoading.value = true; router.replace({ path: `/resources/${row.resource_id}`, query: route.query }); try { detail.value = await resourcePreview(String(row.resource_id)); } finally { detailLoading.value = false; } }
 watch(detailVisible, (visible) => { if (!visible && route.params.resourceId) router.replace({ path: "/resources", query: route.query }); });
 watch([keyword, typeFilter, () => route.query.taskId], resetResourcePage);

frontend/src/views/routes/RoutesView.vue

@@ -6,6 +6,16 @@
       <el-button type="primary" :icon="Plus" @click="createDialog = true">新建航线</el-button>
     </PageHeader>
 
+    <el-alert
+      v-if="bindingTaskId"
+      class="binding-alert"
+      type="info"
+      :closable="false"
+      show-icon
+      title="正在为巡检任务配置航线"
+      :description="`任务 ${bindingTaskId}：请选择${bindingObjectId ? '当前巡检对象的' : ''}已发布航线并点击“绑定到任务”。`"
+    />
+
     <section class="metric-strip">
       <div class="metric-card"><strong>{{ rows.length }}</strong><span>航线</span><small>统一航线台账</small></div>
       <div class="metric-card"><strong>{{ publishedCount }}</strong><span>已发布</span><small>允许任务下发</small></div>
@@ -23,7 +33,7 @@
       <el-table-column label="校验" width="110"><template #default="scope"><el-tag :type="validation(scope.row).valid ? 'success' : 'danger'" effect="plain">{{ validation(scope.row).valid ? '通过' : '未通过' }}</el-tag></template></el-table-column>
       <el-table-column label="状态" width="125"><template #default="scope"><el-tag :type="scope.row.status === 'PUBLISHED' ? 'success' : scope.row.status === 'APPROVED' ? 'warning' : 'info'" effect="plain">{{ routeStatusLabel(scope.row.status) }}</el-tag><small class="cell-subtext">{{ approvalLabel(scope.row.approval_status) }}</small></template></el-table-column>
       <el-table-column label="估算" width="120"><template #default="scope">{{ estimateSummary(scope.row) }}</template></el-table-column>
-      <el-table-column label="操作" width="385" fixed="right"><template #default="scope"><div class="list-actions" @click.stop><el-button link type="primary" @click="validateRow(scope.row)">校验</el-button><el-button link type="primary" @click="estimateRow(scope.row)">估算</el-button><el-button v-if="['DRAFT','REJECTED'].includes(String(scope.row.approval_status || 'DRAFT'))" link type="warning" @click="submitRow(scope.row)">提交审批</el-button><el-button v-if="scope.row.status === 'APPROVED'" link type="success" @click="publishRow(scope.row)">发布</el-button><el-button link @click="newVersion(scope.row)">新版本</el-button><el-button link @click="inspect(scope.row)">详情</el-button></div></template></el-table-column>
+      <el-table-column label="操作" :width="bindingTaskId ? 465 : 385" fixed="right"><template #default="scope"><div class="list-actions" @click.stop><el-button v-if="bindingTaskId && canBindToTask(scope.row)" link type="success" @click="bindToTask(scope.row)">绑定到任务</el-button><el-button link type="primary" @click="validateRow(scope.row)">校验</el-button><el-button link type="primary" @click="estimateRow(scope.row)">估算</el-button><el-button v-if="['DRAFT','REJECTED'].includes(String(scope.row.approval_status || 'DRAFT'))" link type="warning" @click="submitRow(scope.row)">提交审批</el-button><el-button v-if="scope.row.status === 'APPROVED'" link type="success" @click="publishRow(scope.row)">发布</el-button><el-button link @click="newVersion(scope.row)">新版本</el-button><el-button link @click="inspect(scope.row)">详情</el-button></div></template></el-table-column>
     </el-table><el-pagination v-model:current-page="page" v-model:page-size="pageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="filteredRows.length" /></el-card>
 
     <el-dialog v-model="generateDialog" title="自动生成候选航线" width="min(700px, 94vw)" destroy-on-close>
@@ -46,12 +56,48 @@
       <template #footer><el-button @click="createDialog = false">取消</el-button><el-button type="primary" :loading="saving" @click="save">创建并校验</el-button></template>
     </el-dialog>
 
-    <el-drawer v-model="detailVisible" title="航线版本详情" size="760px"><template v-if="selected"><div class="route-visual"><div v-for="(point, index) in parseWaypoints(selected.waypoints)" :key="index" class="route-point" :style="pointStyle(point, index)"><span>{{ index + 1 }}</span></div><div class="route-axis"></div></div><el-descriptions :column="1" border><el-descriptions-item label="航线编号"><span class="entity-id">{{ selected.route_id }}</span></el-descriptions-item><el-descriptions-item label="工作版本">V{{ selected.version_no }} · {{ routeStatusLabel(selected.status) }} · {{ approvalLabel(selected.approval_status) }}</el-descriptions-item><el-descriptions-item label="巡检对象">{{ selected.object_name || '-' }}</el-descriptions-item><el-descriptions-item label="飞行参数">{{ jsonText(selected.flight_parameters) }}</el-descriptions-item><el-descriptions-item label="载荷动作">{{ jsonText(selected.payload_actions) }}</el-descriptions-item><el-descriptions-item label="估算结果">{{ routeEstimateText(selected.estimation_result) }}</el-descriptions-item><el-descriptions-item label="编辑锁">V{{ selected.version_lock ?? 0 }} · {{ selected.edited_by || selected.created_by || '-' }}</el-descriptions-item><el-descriptions-item label="SHA-256"><span class="entity-id">{{ selected.checksum || '-' }}</span></el-descriptions-item></el-descriptions><div class="validation-panel" :class="{ invalid: !validation(selected).valid }"><strong>{{ validation(selected).valid ? '航线校验通过' : '航线校验未通过' }}</strong><p v-for="message in [...(validation(selected).errors || []), ...(validation(selected).warnings || [])]" :key="message">{{ message }}</p><small v-if="!(validation(selected).errors?.length || validation(selected).warnings?.length)">航点数量、高度和速度均符合平台校验规则</small></div><div class="drawer-actions"><el-button type="primary" @click="estimateRow(selected)">重新估算</el-button><el-button :disabled="!['DRAFT','REJECTED'].includes(String(selected.status))" @click="saveWaypointRevision(selected)">保存编辑修订</el-button></div><div class="subsection-head version-head"><div><strong>不可变版本历史</strong><span>回滚将创建一个新草稿，不覆盖历史任务引用</span></div><el-button link @click="loadVersions(selected)">刷新</el-button></div><el-table :data="versionRows" size="small" empty-text="暂无版本"><el-table-column label="版本" width="70"><template #default="scope">V{{ scope.row.version_no }}</template></el-table-column><el-table-column prop="status" label="状态" width="105" /><el-table-column prop="approval_status" label="审批" width="120" /><el-table-column label="估算" min-width="120"><template #default="scope">{{ routeEstimateText(scope.row.estimation_result) }}</template></el-table-column><el-table-column label="摘要" min-width="150"><template #default="scope"><span class="entity-id">{{ String(scope.row.checksum || '-').slice(0, 16) }}</span></template></el-table-column><el-table-column label="操作" width="90"><template #default="scope"><el-button link type="warning" :disabled="scope.row.route_version_id === activeVersionId(selected)" @click="rollbackVersion(scope.row)">回滚</el-button></template></el-table-column></el-table><div class="subsection-head version-head"><div><strong>编辑审计</strong><span>航点和载荷动作修改记录</span></div></div><el-table :data="revisionRows" size="small" empty-text="暂无编辑修订"><el-table-column prop="revision_no" label="修订" width="70" /><el-table-column prop="editor_id" label="编辑人" width="120" /><el-table-column prop="change_summary" label="说明" min-width="160" /><el-table-column label="摘要" min-width="130"><template #default="scope"><span class="entity-id">{{ String(scope.row.checksum || '-').slice(0, 14) }}</span></template></el-table-column></el-table><div class="subsection-head version-head"><div><strong>通用动作 Schema</strong><span>{{ actionSchemas.length }} 项厂商无关动作约束</span></div></div><el-table :data="actionSchemas" size="small"><el-table-column prop="action_type" label="动作" width="150" /><el-table-column prop="device_type" label="设备" width="130" /><el-table-column label="状态" width="90"><template #default="scope"><el-tag size="small" type="success" effect="plain">{{ scope.row.status }}</el-tag></template></el-table-column></el-table></template></el-drawer>
+    <el-drawer v-model="detailVisible" title="航线版本详情" size="820px">
+      <template v-if="selected">
+        <div class="route-visual"><div v-for="(point, index) in editingWaypoints" :key="index" class="route-point" :style="pointStyle(point, index)"><span>{{ index + 1 }}</span></div><div class="route-axis"></div></div>
+        <el-descriptions :column="1" border>
+          <el-descriptions-item label="航线编号"><span class="entity-id">{{ selected.route_id }}</span></el-descriptions-item>
+          <el-descriptions-item label="工作版本">V{{ selected.version_no }} · {{ routeStatusLabel(selected.status) }} · {{ approvalLabel(selected.approval_status) }}</el-descriptions-item>
+          <el-descriptions-item label="巡检对象">{{ selected.object_name || '-' }}</el-descriptions-item>
+          <el-descriptions-item label="载荷动作">{{ jsonText(selected.payload_actions) }}</el-descriptions-item>
+          <el-descriptions-item label="估算结果">{{ routeEstimateText(selected.estimation_result) }}</el-descriptions-item>
+          <el-descriptions-item label="编辑锁">V{{ selected.version_lock ?? 0 }} · {{ selected.edited_by || selected.created_by || '-' }}</el-descriptions-item>
+          <el-descriptions-item label="SHA-256"><span class="entity-id">{{ selected.checksum || '-' }}</span></el-descriptions-item>
+        </el-descriptions>
+
+        <div class="subsection-head version-head"><div><strong>编辑航点与飞行参数</strong><span>{{ immutableSelected ? '保存时自动创建新草稿，不改写已发布版本' : '修改后将形成一条编辑审计记录' }}</span></div><el-button :icon="Plus" @click="addEditingWaypoint">增加航点</el-button></div>
+        <div class="editing-flight-params"><label>默认速度 <el-input-number v-model="editingFlight.speed_mps" :min="1" :max="20" /></label><label>返航高度 <el-input-number v-model="editingFlight.rth_altitude_m" :min="20" :max="500" /></label></div>
+        <el-table :data="editingWaypoints" border size="small">
+          <el-table-column type="index" label="#" width="45" />
+          <el-table-column label="经度" min-width="130"><template #default="scope"><el-input-number v-model="scope.row.longitude" :precision="6" :step="0.001" controls-position="right" /></template></el-table-column>
+          <el-table-column label="纬度" min-width="130"><template #default="scope"><el-input-number v-model="scope.row.latitude" :precision="6" :step="0.001" controls-position="right" /></template></el-table-column>
+          <el-table-column label="高度" min-width="105"><template #default="scope"><el-input-number v-model="scope.row.altitude_m" :min="10" :max="600" controls-position="right" /></template></el-table-column>
+          <el-table-column label="速度" min-width="100"><template #default="scope"><el-input-number v-model="scope.row.speed_mps" :min="1" :max="20" controls-position="right" /></template></el-table-column>
+          <el-table-column label="云台" min-width="100"><template #default="scope"><el-input-number v-model="scope.row.gimbal_pitch_deg" :min="-90" :max="30" controls-position="right" /></template></el-table-column>
+          <el-table-column width="52"><template #default="scope"><el-button link type="danger" :disabled="editingWaypoints.length <= 2" :icon="Delete" @click="editingWaypoints.splice(scope.$index, 1)" /></template></el-table-column>
+        </el-table>
+
+        <div class="validation-panel" :class="{ invalid: !validation(selected).valid }"><strong>{{ validation(selected).valid ? '航线校验通过' : '航线校验未通过' }}</strong><p v-for="message in [...(validation(selected).errors || []), ...(validation(selected).warnings || [])]" :key="message">{{ message }}</p><small v-if="!(validation(selected).errors?.length || validation(selected).warnings?.length)">航点数量、高度和速度均符合平台校验规则</small></div>
+        <div class="drawer-actions"><el-button type="primary" @click="estimateRow(selected)">重新估算</el-button><el-button :loading="saving" :disabled="!editorDirty || editingWaypoints.length < 2" @click="saveWaypointRevision(selected)">{{ immutableSelected ? '另存为新修订' : '保存编辑修订' }}</el-button></div>
+        <div class="subsection-head version-head"><div><strong>不可变版本历史</strong><span>回滚将创建一个新草稿，不覆盖历史任务引用</span></div><el-button link @click="loadVersions(selected)">刷新</el-button></div>
+        <el-table :data="versionRows" size="small" empty-text="暂无版本"><el-table-column label="版本" width="70"><template #default="scope">V{{ scope.row.version_no }}</template></el-table-column><el-table-column prop="status" label="状态" width="105" /><el-table-column prop="approval_status" label="审批" width="120" /><el-table-column label="估算" min-width="120"><template #default="scope">{{ routeEstimateText(scope.row.estimation_result) }}</template></el-table-column><el-table-column label="摘要" min-width="150"><template #default="scope"><span class="entity-id">{{ String(scope.row.checksum || '-').slice(0, 16) }}</span></template></el-table-column><el-table-column label="操作" width="90"><template #default="scope"><el-button link type="warning" :disabled="scope.row.route_version_id === activeVersionId(selected)" @click="rollbackVersion(scope.row)">回滚</el-button></template></el-table-column></el-table>
+        <div class="subsection-head version-head"><div><strong>编辑审计</strong><span>航点和载荷动作修改记录</span></div></div>
+        <el-table :data="revisionRows" size="small" empty-text="暂无编辑修订"><el-table-column prop="revision_no" label="修订" width="70" /><el-table-column prop="editor_id" label="编辑人" width="120" /><el-table-column prop="change_summary" label="说明" min-width="160" /><el-table-column label="摘要" min-width="130"><template #default="scope"><span class="entity-id">{{ String(scope.row.checksum || '-').slice(0, 14) }}</span></template></el-table-column></el-table>
+        <div class="subsection-head version-head"><div><strong>通用动作 Schema</strong><span>{{ actionSchemas.length }} 项厂商无关动作约束</span></div></div>
+        <el-table :data="actionSchemas" size="small"><el-table-column prop="action_type" label="动作" width="150" /><el-table-column prop="device_type" label="设备" width="130" /><el-table-column label="状态" width="90"><template #default="scope"><el-tag size="small" type="success" effect="plain">{{ scope.row.status }}</el-tag></template></el-table-column></el-table>
+      </template>
+    </el-drawer>
   </div>
 </template>
 
 <script setup lang="ts">
 import { computed, onMounted, reactive, ref, watch } from "vue";
+import axios from "axios";
+import { useRoute, useRouter } from "vue-router";
 import { ElMessage } from "element-plus";
 import { Delete, Plus, Refresh, Search } from "@element-plus/icons-vue";
 import PageHeader from "../../components/common/PageHeader.vue";
@@ -59,13 +105,21 @@ import { DEFAULT_PAGE_SIZES, usePagination } from "../../composables/usePaginati
 import { createInspectionRoute, createRouteVersion, estimateRouteVersion, executeWorkflowAction, generateCandidateRoute, inspectionObjects, inspectionRoutes, publishRouteVersion, rollbackRoute, routeActionSchemas, routeEditorRevisions, routeGenerationProfiles, routeVersions, updateRouteWaypoints, validateRouteVersion } from "../../services/api";
 import { safeObject, type Row } from "../../types/demo-run";
 
+const route = useRoute(); const router = useRouter();
 const rows = ref<Row[]>([]); const objects = ref<Row[]>([]); const versionRows = ref<Row[]>([]); const generatorProfiles = ref<Row[]>([]); const revisionRows = ref<Row[]>([]); const actionSchemas = ref<Row[]>([]); const loading = ref(false); const saving = ref(false); const createDialog = ref(false); const generateDialog = ref(false); const detailVisible = ref(false); const selected = ref<Row | null>(null); const keyword = ref(""); const statusFilter = ref("");
+const editingWaypoints = ref<Row[]>([]); const editingFlight = ref<Row>({}); const editingPayloadActions = ref<Row[]>([]); const editorBaseline = ref("");
+const bindingTaskId = computed(() => String(route.query.taskId || ""));
+const bindingObjectId = computed(() => String(route.query.objectId || ""));
 const form = reactive({ name: "铁路沿线巡检航线", object_id: "object-line-demo", speed_mps: 7, rth_altitude_m: 100, waypoints: [{ longitude: 116.09, latitude: 39.095, altitude_m: 80, gimbal_pitch_deg: -45 }, { longitude: 116.125, latitude: 39.115, altitude_m: 80, gimbal_pitch_deg: -45 }, { longitude: 116.165, latitude: 39.105, altitude_m: 80, gimbal_pitch_deg: -45 }] });
 const generateForm = reactive({ object_id: "object-line-demo", generator_type: "", waypoint_count: 5, altitude_m: 80, speed_mps: 7, create_route: true });
-const filteredRows = computed(() => rows.value.filter((row) => (!statusFilter.value || row.status === statusFilter.value) && (!keyword.value || `${row.name} ${row.object_name} ${row.template_name}`.toLowerCase().includes(keyword.value.toLowerCase())))); const publishedCount = computed(() => rows.value.filter((row) => row.status === "PUBLISHED").length); const draftCount = computed(() => rows.value.filter((row) => row.status !== "PUBLISHED").length); const totalWaypoints = computed(() => rows.value.reduce((sum, row) => sum + parseWaypoints(row.waypoints).length, 0));
+const filteredRows = computed(() => rows.value.filter((row) => (!bindingObjectId.value || String(row.object_id) === bindingObjectId.value) && (!statusFilter.value || row.status === statusFilter.value) && (!keyword.value || `${row.name} ${row.object_name} ${row.template_name}`.toLowerCase().includes(keyword.value.toLowerCase())))); const publishedCount = computed(() => rows.value.filter((row) => row.status === "PUBLISHED" || row.route_status === "PUBLISHED").length); const draftCount = computed(() => rows.value.filter((row) => row.status !== "PUBLISHED").length); const totalWaypoints = computed(() => rows.value.reduce((sum, row) => sum + parseWaypoints(row.waypoints).length, 0));
 const generatorOptions = computed(() => generatorProfiles.value.filter((item) => !generateForm.object_id || item.object_type === objects.value.find((object) => object.object_id === generateForm.object_id)?.object_type));
+const immutableSelected = computed(() => !["DRAFT", "REJECTED"].includes(String(selected.value?.status || "")));
+const editorDirty = computed(() => Boolean(selected.value) && editorSnapshot() !== editorBaseline.value);
 const { currentPage: page, pageSize, pagedItems: pagedRows, resetPage } = usePagination(filteredRows);
 function parseWaypoints(value: unknown): Row[] { try { return Array.isArray(value) ? value as Row[] : JSON.parse(String(value || "[]")); } catch { return []; } }
+function editorSnapshot() { return JSON.stringify({ waypoints: editingWaypoints.value.map((point, index) => ({ ...point, sequence: index + 1 })), flight_parameters: editingFlight.value, payload_actions: editingPayloadActions.value }); }
+function resetEditor(row: Row) { editingWaypoints.value = parseWaypoints(row.waypoints).map((point, index) => ({ ...point, sequence: index + 1, speed_mps: Number(point.speed_mps ?? safeObject(row.flight_parameters).speed_mps ?? 6) })); editingFlight.value = { ...safeObject(row.flight_parameters) }; editingPayloadActions.value = parseWaypoints(row.payload_actions).map((item) => ({ ...item })); editorBaseline.value = editorSnapshot(); }
 function validation(row: Row): Row { return safeObject(row.validation_result); }
 function activeVersionId(row: Row) { return String(row.display_version_id || row.working_version_id || row.current_version_id || ""); }
 function routeStatusLabel(value: unknown) { return ({ DRAFT: "草稿", PENDING_APPROVAL: "审批中", APPROVED: "已审批", PUBLISHED: "已发布", REJECTED: "已驳回", RETIRED: "已退役" } as Record<string, string>)[String(value)] || String(value || "-"); }
@@ -77,28 +131,61 @@ function routeEstimate(row: unknown) { return typeof row === "string" ? safeObje
 function routeEstimateText(value: unknown) { const data = routeEstimate(value); return data.estimated_minutes ? `${data.estimated_minutes} 分钟 · 电量 ${data.battery_percent}% · ${data.storage_mb} MB` : "待估算"; }
 function estimateSummary(row: Row) { return routeEstimateText(row.estimation_result); }
 function addWaypoint() { const previous = form.waypoints.at(-1) || { longitude: 116.1, latitude: 39.1, altitude_m: 80, gimbal_pitch_deg: -45 }; form.waypoints.push({ ...previous, longitude: previous.longitude + 0.005, latitude: previous.latitude + 0.003 }); }
-async function inspect(row: Row) { selected.value = row; detailVisible.value = true; await Promise.all([loadVersions(row), loadRevisionRows(row)]); }
+function addEditingWaypoint() { const previous = editingWaypoints.value.at(-1) || { longitude: 116.1, latitude: 39.1, altitude_m: 80, speed_mps: 6, gimbal_pitch_deg: -45 }; editingWaypoints.value.push({ ...previous, sequence: editingWaypoints.value.length + 1, longitude: Number(previous.longitude) + 0.005, latitude: Number(previous.latitude) + 0.003 }); }
-function pointStyle(_point: Row, index: number) { const count = Math.max(2, parseWaypoints(selected.value?.waypoints).length); return { left: `${10 + index * 80 / (count - 1)}%`, top: `${60 - Math.sin(index * 1.4) * 25}%` }; }
+async function inspect(row: Row) { selected.value = row; resetEditor(row); detailVisible.value = true; await Promise.all([loadVersions(row), loadRevisionRows(row)]); }
+function pointStyle(_point: Row, index: number) { const count = Math.max(2, editingWaypoints.value.length); return { left: `${10 + index * 80 / (count - 1)}%`, top: `${60 - Math.sin(index * 1.4) * 25}%` }; }
 async function load() { loading.value = true; try { [rows.value, objects.value, generatorProfiles.value, actionSchemas.value] = await Promise.all([inspectionRoutes(), inspectionObjects(), routeGenerationProfiles(), routeActionSchemas()]); if (!generateForm.generator_type && generatorProfiles.value.length) generateForm.generator_type = String(generatorProfiles.value[0].generator_type); } finally { loading.value = false; } }
 async function save() { if (!form.name || form.waypoints.length < 2) return ElMessage.warning("请填写航线名称并配置至少两个航点"); saving.value = true; try { await createInspectionRoute({ name: form.name, object_id: form.object_id, template_id: "template-railway", waypoints: form.waypoints.map((point, index) => ({ ...point, sequence: index + 1, speed_mps: form.speed_mps, actions: ["TAKE_PHOTO"] })), flight_parameters: { speed_mps: form.speed_mps, rth_altitude_m: form.rth_altitude_m }, payload_actions: [{ type: "TAKE_PHOTO", interval_s: 2 }], created_by: "user-dispatcher" }); createDialog.value = false; ElMessage.success("航线已创建并完成校验"); await load(); } finally { saving.value = false; } }
 async function validateRow(row: Row) { const result = await validateRouteVersion(activeVersionId(row)); ElMessage[result.valid ? "success" : "error"](result.valid ? "航线校验通过" : result.errors.join("；")); await load(); }
 async function estimateRow(row: Row) { const result = await estimateRouteVersion(activeVersionId(row), { created_by: "route-estimator" }); ElMessage[result.warnings?.length ? "warning" : "success"](`估算完成：${result.estimated_minutes} 分钟，电量 ${result.battery_percent}%`); await load(); if (selected.value?.route_id === row.route_id) selected.value = rows.value.find((item) => item.route_id === row.route_id) || selected.value; }
 async function submitRow(row: Row) { const action = String(row.approval_status) === "REJECTED" ? "RESUBMIT" : "SUBMIT"; await executeWorkflowAction("ROUTE_VERSION", activeVersionId(row), action as "SUBMIT" | "RESUBMIT", "航线校验完成，提交发布审批"); ElMessage.success("航线已进入审批队列"); await load(); }
 async function publishRow(row: Row) { try { await publishRouteVersion(activeVersionId(row)); ElMessage.success("航线版本已发布，历史版本已保留"); await load(); } catch { /* global error */ } }
-async function newVersion(row: Row) { const result = await createRouteVersion(String(row.route_id), { source_version_id: activeVersionId(row), change_summary: "基于当前版本创建编辑草稿", created_by: "航线管理员" }); ElMessage.success(`已创建 V${result.version_no} 草稿`); await load(); }
+async function newVersion(row: Row) { const result = await createRouteVersion(String(row.route_id), { source_version_id: activeVersionId(row), change_summary: "基于当前版本创建编辑草稿", created_by: "航线管理员" }); ElMessage.success(`已创建 V${result.version_no} 草稿`); await load(); const refreshed = rows.value.find((item) => item.route_id === row.route_id); if (refreshed && detailVisible.value) await inspect(refreshed); }
 async function loadVersions(row: Row) { versionRows.value = await routeVersions(String(row.route_id)); }
 async function loadRevisionRows(row: Row) { revisionRows.value = await routeEditorRevisions(activeVersionId(row)); }
 async function rollbackVersion(version: Row) { if (!selected.value) return; const result = await rollbackRoute(String(selected.value.route_id), String(version.route_version_id), `回滚至 V${version.version_no}`); ElMessage.success(`已创建回滚草稿 V${result.version_no}，请重新校验和审批`); await load(); const refreshed = rows.value.find((row) => row.route_id === selected.value?.route_id); if (refreshed) { selected.value = refreshed; await loadVersions(refreshed); } }
 async function generateRouteFromObject() { if (!generateForm.object_id) return ElMessage.warning("请选择巡检对象"); saving.value = true; try { const result = await generateCandidateRoute({ object_id: generateForm.object_id, generator_type: generateForm.generator_type || undefined, create_route: generateForm.create_route, route_name: `${objects.value.find((item) => item.object_id === generateForm.object_id)?.name || "对象"} 候选航线`, parameters: { waypoint_count: generateForm.waypoint_count, altitude_m: generateForm.altitude_m, speed_mps: generateForm.speed_mps }, created_by: "route-generator" }); ElMessage.success(generateForm.create_route ? `已创建候选航线 ${result.created_route?.route_id}` : `已生成 ${result.waypoints?.length || 0} 个候选航点`); generateDialog.value = false; await load(); } finally { saving.value = false; } }
-async function saveWaypointRevision(row: Row) { const waypoints = parseWaypoints(row.waypoints).map((point, index) => ({ ...point, sequence: index + 1 })); const result = await updateRouteWaypoints(activeVersionId(row), { waypoints, flight_parameters: safeObject(row.flight_parameters), payload_actions: parseWaypoints(row.payload_actions), change_summary: "前端地图编辑器保存航点修订", editor_id: "route-editor", expected_version_lock: Number(row.version_lock || 0) }); ElMessage.success(`已保存编辑修订 V${result.version_lock}`); await load(); const refreshed = rows.value.find((item) => item.route_id === row.route_id); if (refreshed) { selected.value = refreshed; await Promise.all([loadVersions(refreshed), loadRevisionRows(refreshed)]); } }
+async function saveWaypointRevision(row: Row) {
+  if (!editorDirty.value) return ElMessage.warning("请先修改至少一个航点或飞行参数");
+  if (editingWaypoints.value.length < 2) return ElMessage.warning("航线至少需要两个航点");
+  saving.value = true;
+  try {
+    let targetVersionId = activeVersionId(row); let expectedVersionLock = Number(row.version_lock || 0); let createdVersionNo: number | undefined;
+    if (immutableSelected.value) {
+      const created = await createRouteVersion(String(row.route_id), { source_version_id: activeVersionId(row), change_summary: `基于已发布 V${row.version_no} 编辑航点`, created_by: "航线管理员" });
+      targetVersionId = String(created.route_version_id); expectedVersionLock = 0; createdVersionNo = Number(created.version_no);
+    }
+    const result = await updateRouteWaypoints(targetVersionId, { waypoints: editingWaypoints.value.map((point, index) => ({ ...point, sequence: index + 1 })), flight_parameters: { ...editingFlight.value }, payload_actions: editingPayloadActions.value, change_summary: immutableSelected.value ? "已发布版本航点编辑另存为新修订" : "航点与飞行参数编辑修订", editor_id: "route-editor", expected_version_lock: expectedVersionLock });
+    ElMessage.success(createdVersionNo ? `已创建并保存 V${createdVersionNo} 编辑草稿` : `已保存编辑修订，编辑锁 V${result.version_lock}`);
+    await load(); const refreshed = rows.value.find((item) => item.route_id === row.route_id); if (refreshed) await inspect(refreshed);
+  } finally { saving.value = false; }
+}
+function canBindToTask(row: Row) { return Boolean(bindingTaskId.value) && String(row.route_status || row.status) === "PUBLISHED" && (!bindingObjectId.value || String(row.object_id) === bindingObjectId.value); }
+async function bindToTask(row: Row) {
+  if (!canBindToTask(row)) return ElMessage.warning("请选择与任务巡检对象一致的已发布航线");
+  saving.value = true;
+  try {
+    const response = await axios.put(`/api/v1/inspection/tasks/${encodeURIComponent(bindingTaskId.value)}/route`, { route_id: String(row.route_id), expected_object_id: bindingObjectId.value || undefined, bound_by: "user-dispatcher" });
+    ElMessage.success(`已将航线 ${row.name || row.route_id} 绑定到任务`);
+    await router.push({ path: `/tasks/${bindingTaskId.value}`, query: { routeBound: String(response.data?.data?.route_id || row.route_id) } });
+  } catch (error: any) {
+    ElMessage.error(error?.response?.data?.message || "航线绑定失败，请刷新后重试");
+  } finally { saving.value = false; }
+}
 watch([keyword, statusFilter], resetPage);
 watch(() => generateForm.object_id, () => {
   const first = generatorOptions.value[0];
   if (first) generateForm.generator_type = String(first.generator_type);
 });
-onMounted(load);
+onMounted(async () => {
+  if (bindingObjectId.value) { form.object_id = bindingObjectId.value; generateForm.object_id = bindingObjectId.value; }
+  await load();
+});
 </script>
 
 <style scoped>
 .version-head { margin-top: 24px; }
+.binding-alert { margin-bottom: 16px; }
+.editing-flight-params { display: flex; gap: 24px; align-items: center; margin: 12px 0; }
+.editing-flight-params label { display: flex; gap: 8px; align-items: center; color: var(--el-text-color-regular); }
 </style>

frontend/src/views/tasks/TasksView.vue

@@ -343,7 +343,10 @@ function startManualInspection() { router.push({ path: "/planning", query: { cre
 function openDispatchWorkspace(row: Row) { router.push({ path: "/uav-operations", query: { tab: "missions", dispatchTaskId: String(row.task_id) } }); }
 function openMission(row: Row) { router.push({ path: "/uav-operations", query: { tab: "missions", missionId: String(row.mission_id) } }); }
 function openResources(row: Row) { router.push({ path: "/resources", query: { taskId: String(row.task_id) } }); }
-function openMap(row: Row) { router.push({ path: "/gis", query: { taskId: String(row.task_id) } }); }
+function openMap(row: Row) {
+  const objectId = parseIds(row.object_scope)[0];
+  router.push({ path: "/gis", query: objectId ? { taskId: String(row.task_id), objectId } : { taskId: String(row.task_id) } });
+}
 function configureRoute(row: Row) {
   const objectId = parseIds(row.object_scope)[0];
   router.push({ path: "/routes", query: objectId ? { objectId, taskId: String(row.task_id) } : { taskId: String(row.task_id) } });

frontend/src/views/uav-operations/UavOperationsView.vue

@@ -63,7 +63,7 @@
           <el-table-column prop="vendor_code" label="厂商" width="95" />
           <el-table-column label="进度" min-width="140"><template #default="scope"><el-progress :percentage="scope.row.progress" :stroke-width="8" /></template></el-table-column>
           <el-table-column label="状态" width="105"><template #default="scope"><el-tag :type="missionTag(scope.row.status)" effect="plain">{{ missionLabel(scope.row.status) }}</el-tag></template></el-table-column>
-          <el-table-column label="操作" width="330" fixed="right"><template #default="scope"><div class="list-actions" @click.stop><el-button v-if="scope.row.status === 'DISPATCHED' && scope.row.vendor_code === 'SIMULATOR'" link type="primary" @click="command(scope.row, 'MISSION_START')">开始</el-button><el-button v-if="scope.row.status === 'FLYING'" link type="warning" @click="command(scope.row, 'MISSION_PAUSE')">暂停</el-button><el-button v-if="scope.row.status === 'PAUSED'" link type="primary" @click="command(scope.row, 'MISSION_RESUME')">恢复</el-button><el-button v-if="scope.row.status === 'FLYING' && scope.row.vendor_code === 'SIMULATOR'" link type="primary" @click="command(scope.row, 'SIMULATE_PROGRESS')">推进</el-button><el-button v-if="['FLYING','PAUSED'].includes(scope.row.status)" link type="danger" @click="command(scope.row, 'RETURN_HOME')">返航</el-button><el-button v-if="['DISPATCHED','PREPARING'].includes(scope.row.status)" link type="danger" @click="command(scope.row, 'MISSION_CANCEL')">取消</el-button><el-button link @click="inspectMission(scope.row)">遥测</el-button></div></template></el-table-column>
+          <el-table-column label="操作" width="350" fixed="right"><template #default="scope"><div class="list-actions" @click.stop><el-button v-if="scope.row.status === 'DISPATCHED' && scope.row.vendor_code === 'SIMULATOR'" link type="primary" @click="command(scope.row, 'MISSION_START')">开始</el-button><el-button v-if="scope.row.status === 'FLYING'" link type="warning" @click="command(scope.row, 'MISSION_PAUSE')">暂停</el-button><el-button v-if="scope.row.status === 'PAUSED'" link type="primary" @click="command(scope.row, 'MISSION_RESUME')">恢复</el-button><el-button v-if="['FLYING','RETURNING'].includes(scope.row.status) && scope.row.vendor_code === 'SIMULATOR'" link type="primary" @click="command(scope.row, 'SIMULATE_PROGRESS')">{{ scope.row.status === 'RETURNING' ? '完成返航' : '推进' }}</el-button><el-button v-if="['FLYING','PAUSED'].includes(scope.row.status)" link type="danger" @click="command(scope.row, 'RETURN_HOME')">返航</el-button><el-button v-if="['DISPATCHED','PREPARING'].includes(scope.row.status)" link type="danger" @click="command(scope.row, 'MISSION_CANCEL')">取消</el-button><el-button link @click="inspectMission(scope.row)">遥测</el-button></div></template></el-table-column>
         </el-table><el-pagination v-model:current-page="missionPage" v-model:page-size="missionPageSize" class="table-pagination" background layout="total, sizes, prev, pager, next" :page-sizes="DEFAULT_PAGE_SIZES" :total="missions.length" /></el-card>
       </el-tab-pane>
     </el-tabs>

frontend/src/views/visualization-demo/VideoAiDemoView.vue

@@ -0,0 +1,635 @@
+<template>
+  <div class="video-demo-page">
+    <PageHeader title="视频 AI 可视化演示" description="本地视频播放、实时检测分割、完整结果视频导出">
+      <input ref="fileInput" class="hidden-file-input" type="file" accept="video/*" @change="onVideoSelected" />
+      <el-button :icon="UploadFilled" type="primary" @click="fileInput?.click()">上传视频</el-button>
+      <el-button :icon="Refresh" :loading="capabilityLoading" @click="loadCapabilities">刷新运行时</el-button>
+      <el-button :icon="Cpu" :loading="warming" @click="warmupModels">模型预热</el-button>
+    </PageHeader>
+
+    <section class="video-demo-layout">
+      <main class="video-workspace">
+        <div ref="stageRef" class="video-stage">
+          <video
+            ref="videoRef"
+            :src="videoUrl"
+            controls
+            playsinline
+            @loadedmetadata="onVideoReady"
+            @play="startFrameLoop"
+            @pause="stopFrameLoop"
+            @seeked="clearOverlay"
+          ></video>
+          <canvas ref="overlayRef" class="video-overlay"></canvas>
+          <div v-if="!videoUrl" class="video-empty">
+            <el-icon><VideoPlay /></el-icon>
+            <strong>等待视频文件</strong>
+          </div>
+        </div>
+
+        <section class="video-result-strip">
+          <div><span>当前时间</span><strong>{{ currentTimeText }}</strong></div>
+          <div><span>检测目标</span><strong>{{ latestResult?.results.detections.length || 0 }}</strong></div>
+          <div><span>分割区域</span><strong>{{ latestResult?.results.segments.length || 0 }}</strong></div>
+          <div><span>端到端延迟</span><strong>{{ latencyText }}</strong></div>
+          <div><span>Provider</span><strong>{{ providerText }}</strong></div>
+        </section>
+
+        <el-card class="workspace-card result-card" shadow="never">
+          <template #header>
+            <div class="card-heading">
+              <strong>当前帧结果</strong>
+              <span>{{ realtimeWarnings.length ? `${realtimeWarnings.length} 条提示` : "实时叠加" }}</span>
+            </div>
+          </template>
+          <el-alert v-for="item in realtimeWarnings" :key="`${item.code}-${item.model_group}`" :title="item.message" type="warning" show-icon :closable="false" />
+          <el-table :data="resultRows" size="small" max-height="260" empty-text="暂无结果">
+            <el-table-column prop="type" label="类型" width="86" />
+            <el-table-column prop="category" label="类别" min-width="130" show-overflow-tooltip />
+            <el-table-column prop="confidence" label="置信度" width="90" />
+            <el-table-column prop="model" label="模型" min-width="150" show-overflow-tooltip />
+          </el-table>
+        </el-card>
+      </main>
+
+      <aside class="video-control-panel">
+        <el-card class="workspace-card" shadow="never">
+          <template #header><div class="card-heading"><strong>推理控制</strong><span>{{ videoFile?.name || "未选择视频" }}</span></div></template>
+          <div class="control-stack">
+            <label class="switch-row"><span><strong>目标检测</strong><small>检测框、类别、置信度</small></span><el-switch v-model="detectEnabled" @change="onModeChanged" /></label>
+            <label class="switch-row"><span><strong>图像分割</strong><small>掩膜、轮廓、面积占比</small></span><el-switch v-model="segmentEnabled" @change="onModeChanged" /></label>
+            <div class="form-row">
+              <span>检测场景</span>
+              <el-select v-model="selectedDetectionScene" @change="onSceneChanged">
+                <el-option v-for="scene in detectionScenes" :key="scene.id" :label="scene.label" :value="scene.id" />
+              </el-select>
+            </div>
+            <div class="slider-field"><span>检测阈值 {{ confidenceThreshold.toFixed(2) }}</span><el-slider v-model="confidenceThreshold" :min="0.05" :max="0.95" :step="0.05" /></div>
+            <div class="slider-field"><span>分割阈值 {{ maskThreshold.toFixed(2) }}</span><el-slider v-model="maskThreshold" :min="0.1" :max="0.9" :step="0.05" /></div>
+            <div class="form-row"><span>推理宽度</span><el-select v-model="maxInferenceWidth"><el-option v-for="item in widthOptions" :key="item" :label="`${item}px`" :value="item" /></el-select></div>
+            <div class="form-row"><span>检测 FPS</span><el-input-number v-model="detectionFps" :min="1" :max="12" :step="1" controls-position="right" /></div>
+            <div class="form-row"><span>分割 FPS</span><el-input-number v-model="segmentationFps" :min="1" :max="6" :step="1" controls-position="right" /></div>
+          </div>
+        </el-card>
+
+        <el-card class="workspace-card" shadow="never">
+          <template #header><div class="card-heading"><strong>运行时</strong><span>{{ runtimeReadyText }}</span></div></template>
+          <div class="runtime-list">
+            <div><span>GPU</span><strong>{{ capabilities?.gpu?.name || "未检测" }}</strong></div>
+            <div><span>执行器</span><strong>{{ capabilities?.execution_provider || "-" }}</strong></div>
+            <div><span>模型</span><strong>{{ readyModelCount }}/{{ capabilities?.models?.length || 0 }} 就绪</strong></div>
+            <div><span>输出根目录</span><strong>{{ capabilities?.export?.output_root || "-" }}</strong></div>
+          </div>
+        </el-card>
+
+        <el-card class="workspace-card" shadow="never">
+          <template #header><div class="card-heading"><strong>完整结果视频</strong><span>{{ exportStatusText }}</span></div></template>
+          <div class="control-stack">
+            <div class="form-row"><span>分析步长</span><el-input-number v-model="analysisStride" :min="1" :max="10" controls-position="right" /></div>
+            <label class="switch-row compact"><span><strong>复用结果</strong><small>非推理帧沿用上一帧</small></span><el-switch v-model="reuseLastResult" /></label>
+            <el-button type="primary" :icon="VideoPlay" :disabled="!canExport" :loading="exportStarting" @click="startExport">生成结果视频</el-button>
+            <el-progress v-if="exportJob" :percentage="Math.round(exportJob.progress?.percent || 0)" :status="exportJob.status === 'failed' ? 'exception' : exportJob.status === 'succeeded' ? 'success' : undefined" />
+            <div v-if="exportJob" class="export-meta">
+              <span>{{ exportProgressText }}</span>
+              <span v-if="exportJob.error">{{ exportJob.error }}</span>
+              <span v-if="exportOutputDir"><el-icon><FolderOpened /></el-icon>{{ exportOutputDir }}</span>
+            </div>
+            <div v-if="exportJob?.status === 'succeeded'" class="export-actions">
+              <el-button :icon="VideoPlay" @click="previewExportVideo">预览</el-button>
+              <el-button :icon="Download" @click="downloadExportFile('annotated.mp4')">视频</el-button>
+              <el-button :icon="Download" @click="downloadExportFile('results.json')">JSON</el-button>
+              <el-button :icon="Download" @click="downloadExportFile('run-metadata.json')">元数据</el-button>
+            </div>
+          </div>
+        </el-card>
+      </aside>
+    </section>
+
+    <el-dialog v-model="previewVisible" title="结果视频预览" width="860px" destroy-on-close>
+      <video v-if="exportJob" class="export-preview-video" :src="videoExportFileUrl(exportJob.run_id, 'annotated.mp4')" controls autoplay></video>
+    </el-dialog>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { computed, nextTick, onBeforeUnmount, onMounted, ref, watch } from "vue";
+import { Cpu, Download, FolderOpened, Refresh, UploadFilled, VideoPlay } from "@element-plus/icons-vue";
+import { ElMessage } from "element-plus";
+import PageHeader from "../../components/common/PageHeader.vue";
+import {
+  createVideoExportJob,
+  inferVideoFrame,
+  videoDemoCapabilities,
+  videoExportFileUrl,
+  videoExportJob,
+  warmupVideoDemoModels,
+  type FrameInferenceResponse,
+  type VideoDemoWarning,
+  type VideoExportJob,
+  type VideoMaskRle
+} from "../../services/videoDemoApi";
+
+const fileInput = ref<HTMLInputElement>();
+const stageRef = ref<HTMLElement>();
+const videoRef = ref<HTMLVideoElement>();
+const overlayRef = ref<HTMLCanvasElement>();
+const videoFile = ref<File>();
+const videoUrl = ref("");
+const capabilities = ref<any>();
+const capabilityLoading = ref(false);
+const warming = ref(false);
+const detectEnabled = ref(false);
+const segmentEnabled = ref(false);
+const confidenceThreshold = ref(0.45);
+const maskThreshold = ref(0.5);
+const maxInferenceWidth = ref(960);
+const detectionFps = ref(8);
+const segmentationFps = ref(3);
+const latestResult = ref<FrameInferenceResponse>();
+const realtimeWarnings = ref<VideoDemoWarning[]>([]);
+const pendingFrame = ref(false);
+const currentTime = ref(0);
+const exportStarting = ref(false);
+const exportJob = ref<VideoExportJob>();
+const exportOutputDir = ref("");
+const analysisStride = ref(1);
+const reuseLastResult = ref(true);
+const previewVisible = ref(false);
+let frameTimer: number | undefined;
+let exportTimer: number | undefined;
+let resizeObserver: ResizeObserver | undefined;
+const captureCanvas = document.createElement("canvas");
+const widthOptions = [640, 960, 1280];
+const selectedDetectionScene = ref("traffic-driving");
+const fallbackDetectionScenes = [
+  {
+    id: "inspection",
+    label: "铁路/无人机巡检",
+    detection_model_version: undefined,
+    segmentation_model_version: undefined,
+    default_confidence_threshold: 0.45
+  },
+  {
+    id: "traffic-driving",
+    label: "驾车/道路交通",
+    detection_model_version: "traffic-yolov8n-coco",
+    segmentation_model_version: "traffic-yolov8n-seg-coco",
+    default_confidence_threshold: 0.35
+  }
+];
+
+const canExport = computed(() => Boolean(videoFile.value && (detectEnabled.value || segmentEnabled.value) && exportJob.value?.status !== "running" && exportJob.value?.status !== "queued"));
+const detectionScenes = computed(() => capabilities.value?.detection_scenes?.length ? capabilities.value.detection_scenes : fallbackDetectionScenes);
+const selectedScene = computed(() => detectionScenes.value.find((item: any) => item.id === selectedDetectionScene.value) || detectionScenes.value[0]);
+const selectedDetectionModelVersion = computed(() => selectedScene.value?.detection_model_version || undefined);
+const selectedSegmentationModelVersion = computed(() => selectedScene.value?.segmentation_model_version || undefined);
+const readyModelCount = computed(() => (capabilities.value?.models || []).filter((item: any) => item.artifact_installed).length);
+const providerText = computed(() => String(latestResult.value?.runtime.provider || capabilities.value?.execution_provider || "-"));
+const runtimeReadyText = computed(() => capabilities.value?.accelerated ? "GPU 就绪" : "CPU 或未就绪");
+const latencyText = computed(() => latestResult.value ? `${latestResult.value.runtime.total_latency_ms} ms` : "-");
+const currentTimeText = computed(() => `${currentTime.value.toFixed(2)} s`);
+const exportStatusText = computed(() => {
+  if (!exportJob.value) return "未开始";
+  return ({ queued: "排队中", running: "生成中", succeeded: "已完成", failed: "失败" } as Record<string, string>)[exportJob.value.status] || exportJob.value.status;
+});
+const exportProgressText = computed(() => {
+  if (!exportJob.value) return "";
+  const progress = exportJob.value.progress || { processed_frames: 0, total_frames: 0, percent: 0 };
+  const total = progress.total_frames || "-";
+  const eta = progress.eta_seconds ? `，剩余 ${Math.round(progress.eta_seconds)}s` : "";
+  return `${progress.processed_frames}/${total} 帧${eta}`;
+});
+const resultRows = computed(() => {
+  const detections = (latestResult.value?.results.detections || []).map((item) => ({
+    type: "检测",
+    category: item.category,
+    confidence: item.confidence.toFixed(2),
+    model: `${item.model_group} ${item.model_version || ""}`
+  }));
+  const segments = (latestResult.value?.results.segments || []).map((item) => ({
+    type: "分割",
+    category: item.category,
+    confidence: item.confidence.toFixed(2),
+    model: `${item.model_group} ${item.model_version || ""}`
+  }));
+  return [...detections, ...segments];
+});
+
+async function loadCapabilities() {
+  capabilityLoading.value = true;
+  try {
+    capabilities.value = await videoDemoCapabilities();
+    maxInferenceWidth.value = capabilities.value?.recommended?.max_inference_width || maxInferenceWidth.value;
+    detectionFps.value = capabilities.value?.recommended?.detection_fps || detectionFps.value;
+    segmentationFps.value = capabilities.value?.recommended?.segmentation_fps || segmentationFps.value;
+    const scenes = capabilities.value?.detection_scenes || [];
+    if (scenes.length && !scenes.some((item: any) => item.id === selectedDetectionScene.value)) {
+      selectedDetectionScene.value = scenes[0].id;
+    }
+  } catch (error) {
+    ElMessage.warning(errorMessage(error));
+  } finally {
+    capabilityLoading.value = false;
+  }
+}
+
+async function warmupModels() {
+  warming.value = true;
+  try {
+    const models = [detectEnabled.value && "vision-detector", segmentEnabled.value && "vision-segmenter"].filter(Boolean) as string[];
+    const warmupTargets = models.length ? models : ["vision-detector", "vision-segmenter"];
+    await warmupVideoDemoModels(warmupTargets, {
+      "vision-detector": selectedDetectionModelVersion.value,
+      "vision-segmenter": selectedSegmentationModelVersion.value
+    });
+    await loadCapabilities();
+    ElMessage.success("模型预热完成");
+  } catch (error) {
+    ElMessage.warning(errorMessage(error));
+  } finally {
+    warming.value = false;
+  }
+}
+
+function onVideoSelected(event: Event) {
+  const file = (event.target as HTMLInputElement).files?.[0];
+  if (!file) return;
+  if (videoUrl.value) URL.revokeObjectURL(videoUrl.value);
+  videoFile.value = file;
+  videoUrl.value = URL.createObjectURL(file);
+  latestResult.value = undefined;
+  realtimeWarnings.value = [];
+  exportJob.value = undefined;
+  exportOutputDir.value = "";
+  void nextTick(syncOverlaySize);
+}
+
+function onVideoReady() {
+  syncOverlaySize();
+  drawOverlay();
+}
+
+function onModeChanged() {
+  clearOverlay();
+  if (detectEnabled.value || segmentEnabled.value) startFrameLoop();
+  else stopFrameLoop();
+}
+
+function onSceneChanged() {
+  const defaultThreshold = Number(selectedScene.value?.default_confidence_threshold);
+  if (Number.isFinite(defaultThreshold)) confidenceThreshold.value = defaultThreshold;
+  clearOverlay();
+  if (detectEnabled.value || segmentEnabled.value) startFrameLoop();
+}
+
+function startFrameLoop() {
+  window.clearTimeout(frameTimer);
+  const video = videoRef.value;
+  if (!video || video.paused || (!detectEnabled.value && !segmentEnabled.value)) return;
+  frameTimer = window.setTimeout(captureAndInferFrame, frameIntervalMs());
+}
+
+function stopFrameLoop() {
+  window.clearTimeout(frameTimer);
+}
+
+async function captureAndInferFrame() {
+  const video = videoRef.value;
+  if (!video || video.paused || pendingFrame.value || (!detectEnabled.value && !segmentEnabled.value)) {
+    startFrameLoop();
+    return;
+  }
+  pendingFrame.value = true;
+  try {
+    const blob = await captureCurrentFrame(video);
+    const form = new FormData();
+    form.append("frame", blob, "frame.jpg");
+    form.append("session_id", sessionId());
+    form.append("timestamp_ms", String(video.currentTime * 1000));
+    form.append("source_width", String(video.videoWidth));
+    form.append("source_height", String(video.videoHeight));
+    form.append("detect_enabled", String(detectEnabled.value));
+    form.append("segment_enabled", String(segmentEnabled.value));
+    form.append("confidence_threshold", String(confidenceThreshold.value));
+    form.append("mask_threshold", String(maskThreshold.value));
+    form.append("max_detections", "100");
+    form.append("max_inference_width", String(maxInferenceWidth.value));
+    form.append("detection_scene", selectedDetectionScene.value);
+    if (selectedDetectionModelVersion.value) form.append("detection_model_version", selectedDetectionModelVersion.value);
+    if (selectedSegmentationModelVersion.value) form.append("segmentation_model_version", selectedSegmentationModelVersion.value);
+    const result = await inferVideoFrame(form);
+    latestResult.value = result;
+    realtimeWarnings.value = result.warnings || [];
+    currentTime.value = video.currentTime;
+    drawOverlay();
+  } catch (error) {
+    realtimeWarnings.value = [{ code: "FRAME_INFERENCE_FAILED", message: errorMessage(error) }];
+  } finally {
+    pendingFrame.value = false;
+    startFrameLoop();
+  }
+}
+
+function captureCurrentFrame(video: HTMLVideoElement): Promise<Blob> {
+  const scale = Math.min(1, maxInferenceWidth.value / Math.max(1, video.videoWidth));
+  captureCanvas.width = Math.max(1, Math.round(video.videoWidth * scale));
+  captureCanvas.height = Math.max(1, Math.round(video.videoHeight * scale));
+  const context = captureCanvas.getContext("2d");
+  if (!context) throw new Error("Canvas 不可用");
+  context.drawImage(video, 0, 0, captureCanvas.width, captureCanvas.height);
+  return new Promise((resolve, reject) => {
+    captureCanvas.toBlob((blob) => blob ? resolve(blob) : reject(new Error("无法生成视频帧")), "image/jpeg", 0.82);
+  });
+}
+
+function frameIntervalMs() {
+  const fps = segmentEnabled.value ? Math.min(detectionFps.value, segmentationFps.value) : detectionFps.value;
+  return Math.max(80, Math.round(1000 / Math.max(1, fps)));
+}
+
+async function startExport() {
+  if (!videoFile.value) return;
+  exportStarting.value = true;
+  try {
+    const form = new FormData();
+    form.append("video", videoFile.value);
+    form.append("detect_enabled", String(detectEnabled.value));
+    form.append("segment_enabled", String(segmentEnabled.value));
+    form.append("confidence_threshold", String(confidenceThreshold.value));
+    form.append("mask_threshold", String(maskThreshold.value));
+    form.append("max_inference_width", String(maxInferenceWidth.value));
+    form.append("analysis_stride", String(analysisStride.value));
+    form.append("reuse_last_result", String(reuseLastResult.value));
+    form.append("max_detections", "100");
+    form.append("detection_scene", selectedDetectionScene.value);
+    if (selectedDetectionModelVersion.value) form.append("detection_model_version", selectedDetectionModelVersion.value);
+    if (selectedSegmentationModelVersion.value) form.append("segmentation_model_version", selectedSegmentationModelVersion.value);
+    const created = await createVideoExportJob(form);
+    exportOutputDir.value = created.output_dir;
+    exportJob.value = { ...created, progress: { processed_frames: 0, percent: 0 }, outputs: {}, warnings: [] } as VideoExportJob;
+    pollExportJob(created.run_id);
+  } catch (error) {
+    ElMessage.error(errorMessage(error));
+  } finally {
+    exportStarting.value = false;
+  }
+}
+
+async function pollExportJob(runId: string) {
+  window.clearTimeout(exportTimer);
+  try {
+    exportJob.value = await videoExportJob(runId);
+    if (exportJob.value.outputs?.annotated_video) exportOutputDir.value = exportJob.value.outputs.annotated_video.replace(/[/\\]annotated\.mp4$/, "");
+    if (["queued", "running"].includes(exportJob.value.status)) {
+      exportTimer = window.setTimeout(() => pollExportJob(runId), 1200);
+    }
+  } catch (error) {
+    ElMessage.warning(errorMessage(error));
+  }
+}
+
+function previewExportVideo() {
+  previewVisible.value = true;
+}
+
+function downloadExportFile(fileName: string) {
+  if (!exportJob.value) return;
+  window.open(videoExportFileUrl(exportJob.value.run_id, fileName), "_blank", "noopener,noreferrer");
+}
+
+function syncOverlaySize() {
+  const canvas = overlayRef.value;
+  const stage = stageRef.value;
+  if (!canvas || !stage) return;
+  const rect = stage.getBoundingClientRect();
+  const dpr = window.devicePixelRatio || 1;
+  canvas.width = Math.max(1, Math.round(rect.width * dpr));
+  canvas.height = Math.max(1, Math.round(rect.height * dpr));
+  canvas.style.width = `${rect.width}px`;
+  canvas.style.height = `${rect.height}px`;
+  drawOverlay();
+}
+
+function drawOverlay() {
+  const canvas = overlayRef.value;
+  const stage = stageRef.value;
+  const video = videoRef.value;
+  if (!canvas || !stage || !video) return;
+  const rect = stage.getBoundingClientRect();
+  const dpr = window.devicePixelRatio || 1;
+  const context = canvas.getContext("2d");
+  if (!context) return;
+  context.setTransform(dpr, 0, 0, dpr, 0, 0);
+  context.clearRect(0, 0, rect.width, rect.height);
+  if (!latestResult.value || !video.videoWidth || !video.videoHeight) return;
+  const box = videoDrawBox(rect.width, rect.height, video.videoWidth, video.videoHeight);
+  for (const segment of latestResult.value.results.segments) {
+    if (segment.mask) drawSegmentMask(context, box, segment.mask, segment.category);
+    else drawSegment(context, box, segment.polygon, segment.category);
+  }
+  for (const detection of latestResult.value.results.detections) drawDetection(context, box, detection.bbox, detection.category, detection.confidence, colorFor(detection.category));
+}
+
+function clearOverlay() {
+  latestResult.value = undefined;
+  realtimeWarnings.value = [];
+  const canvas = overlayRef.value;
+  const context = canvas?.getContext("2d");
+  if (canvas && context) context.clearRect(0, 0, canvas.width, canvas.height);
+}
+
+function videoDrawBox(stageWidth: number, stageHeight: number, sourceWidth: number, sourceHeight: number) {
+  const stageRatio = stageWidth / stageHeight;
+  const videoRatio = sourceWidth / sourceHeight;
+  if (videoRatio > stageRatio) {
+    const width = stageWidth;
+    const height = width / videoRatio;
+    return { x: 0, y: (stageHeight - height) / 2, width, height };
+  }
+  const height = stageHeight;
+  const width = height * videoRatio;
+  return { x: (stageWidth - width) / 2, y: 0, width, height };
+}
+
+function drawSegment(context: CanvasRenderingContext2D, box: any, polygon: number[][], category: string) {
+  if (!polygon?.length) return;
+  const color = colorFor(category);
+  context.beginPath();
+  polygon.forEach((point, index) => {
+    const x = box.x + point[0] * box.width;
+    const y = box.y + point[1] * box.height;
+    index === 0 ? context.moveTo(x, y) : context.lineTo(x, y);
+  });
+  context.closePath();
+  context.fillStyle = alphaColorFor(category, 0.33);
+  context.strokeStyle = color;
+  context.lineWidth = 2;
+  context.fill();
+  context.stroke();
+}
+
+function drawSegmentMask(context: CanvasRenderingContext2D, box: any, mask: VideoMaskRle, category: string) {
+  if (mask.encoding !== "rle" || mask.width <= 0 || mask.height <= 0) return;
+  const decoded = decodeRleMask(mask);
+  const maskCanvas = document.createElement("canvas");
+  maskCanvas.width = mask.width;
+  maskCanvas.height = mask.height;
+  const maskContext = maskCanvas.getContext("2d");
+  if (!maskContext) return;
+  const imageData = maskContext.createImageData(mask.width, mask.height);
+  const { r, g, b } = rgbFor(category);
+  for (let index = 0; index < decoded.length; index += 1) {
+    if (!decoded[index]) continue;
+    const offset = index * 4;
+    imageData.data[offset] = r;
+    imageData.data[offset + 1] = g;
+    imageData.data[offset + 2] = b;
+    imageData.data[offset + 3] = 92;
+  }
+  maskContext.putImageData(imageData, 0, 0);
+  context.save();
+  context.imageSmoothingEnabled = false;
+  context.drawImage(maskCanvas, box.x, box.y, box.width, box.height);
+  context.restore();
+}
+
+function decodeRleMask(mask: VideoMaskRle) {
+  const total = Math.max(0, Math.floor(mask.width * mask.height));
+  const decoded = new Uint8Array(total);
+  let offset = 0;
+  let value = 0;
+  for (const count of mask.counts) {
+    const length = Math.max(0, Math.floor(Number(count) || 0));
+    const end = Math.min(total, offset + length);
+    if (value) decoded.fill(1, offset, end);
+    offset = end;
+    if (offset >= total) break;
+    value = value ? 0 : 1;
+  }
+  return decoded;
+}
+
+function drawDetection(context: CanvasRenderingContext2D, box: any, bbox: number[], label: string, confidence: number, color: string) {
+  if (bbox.length !== 4) return;
+  const x = box.x + bbox[0] * box.width;
+  const y = box.y + bbox[1] * box.height;
+  const width = (bbox[2] - bbox[0]) * box.width;
+  const height = (bbox[3] - bbox[1]) * box.height;
+  context.strokeStyle = color;
+  context.lineWidth = 2;
+  context.strokeRect(x, y, width, height);
+  const text = `${label} ${confidence.toFixed(2)}`;
+  context.font = "12px Microsoft YaHei, sans-serif";
+  const textWidth = context.measureText(text).width + 10;
+  context.fillStyle = color;
+  context.fillRect(x, Math.max(0, y - 22), textWidth, 20);
+  context.fillStyle = "#fff";
+  context.fillText(text, x + 5, Math.max(14, y - 8));
+}
+
+function hueFor(value: string) {
+  let hash = 0;
+  for (let index = 0; index < value.length; index += 1) hash = ((hash << 5) - hash) + value.charCodeAt(index);
+  return Math.abs(hash) % 360;
+}
+
+function colorFor(value: string) {
+  return `hsl(${hueFor(value)}, 72%, 48%)`;
+}
+
+function alphaColorFor(value: string, alpha: number) {
+  return `hsla(${hueFor(value)}, 72%, 48%, ${alpha})`;
+}
+
+function rgbFor(value: string) {
+  const hue = hueFor(value) / 360;
+  const saturation: number = 0.72;
+  const lightness: number = 0.48;
+  if (saturation === 0) {
+    const gray = Math.round(lightness * 255);
+    return { r: gray, g: gray, b: gray };
+  }
+  const q = lightness < 0.5 ? lightness * (1 + saturation) : lightness + saturation - lightness * saturation;
+  const p = 2 * lightness - q;
+  const toRgb = (channel: number) => {
+    let t = channel;
+    if (t < 0) t += 1;
+    if (t > 1) t -= 1;
+    if (t < 1 / 6) return p + (q - p) * 6 * t;
+    if (t < 1 / 2) return q;
+    if (t < 2 / 3) return p + (q - p) * (2 / 3 - t) * 6;
+    return p;
+  };
+  return {
+    r: Math.round(toRgb(hue + 1 / 3) * 255),
+    g: Math.round(toRgb(hue) * 255),
+    b: Math.round(toRgb(hue - 1 / 3) * 255)
+  };
+}
+
+function sessionId() {
+  return `video-demo-${videoFile.value?.name || "local"}`;
+}
+
+function errorMessage(error: unknown) {
+  const item = error as { response?: { data?: { detail?: string; message?: string } }; message?: string };
+  return item.response?.data?.detail || item.response?.data?.message || item.message || "操作失败";
+}
+
+onMounted(() => {
+  void loadCapabilities();
+  resizeObserver = new ResizeObserver(syncOverlaySize);
+  if (stageRef.value) resizeObserver.observe(stageRef.value);
+  window.addEventListener("resize", syncOverlaySize);
+});
+
+watch([confidenceThreshold, maskThreshold, maxInferenceWidth], clearOverlay);
+
+onBeforeUnmount(() => {
+  stopFrameLoop();
+  window.clearTimeout(exportTimer);
+  window.removeEventListener("resize", syncOverlaySize);
+  resizeObserver?.disconnect();
+  if (videoUrl.value) URL.revokeObjectURL(videoUrl.value);
+});
+</script>
+
+<style scoped>
+.video-demo-layout { display: grid; grid-template-columns: minmax(0, 1fr) 360px; gap: 16px; align-items: start; }
+.video-workspace, .video-control-panel { min-width: 0; display: grid; gap: 14px; }
+.video-stage { position: relative; min-height: 520px; overflow: hidden; border: 1px solid #d5dee9; border-radius: 6px; background: #111827; }
+.video-stage video, .video-overlay { position: absolute; inset: 0; width: 100%; height: 100%; }
+.video-stage video { object-fit: contain; background: #111827; }
+.video-overlay { z-index: 2; pointer-events: none; }
+.video-empty { position: absolute; inset: 0; z-index: 3; display: grid; place-items: center; align-content: center; gap: 10px; color: #dbeafe; background: linear-gradient(135deg, #111827, #1f2937); }
+.video-empty .el-icon { font-size: 46px; }
+.video-result-strip { display: grid; grid-template-columns: repeat(5, minmax(0, 1fr)); gap: 10px; }
+.video-result-strip > div, .runtime-list > div { min-width: 0; padding: 12px; border: 1px solid #dce4ed; border-radius: 6px; background: #fff; }
+.video-result-strip span, .video-result-strip strong, .runtime-list span, .runtime-list strong { display: block; min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+.video-result-strip span, .runtime-list span { color: #64748b; font-size: 11px; }
+.video-result-strip strong, .runtime-list strong { margin-top: 6px; color: #17365d; font-size: 15px; }
+.result-card :deep(.el-alert) { margin-bottom: 8px; }
+.control-stack { display: grid; gap: 14px; }
+.switch-row { min-height: 58px; padding: 10px 11px; display: flex; align-items: center; justify-content: space-between; gap: 12px; border: 1px solid #e2e8f0; border-radius: 6px; background: #f8fafc; }
+.switch-row.compact { min-height: 50px; }
+.switch-row span strong, .switch-row span small { display: block; }
+.switch-row span strong { color: #17365d; font-size: 13px; }
+.switch-row span small { margin-top: 4px; color: #64748b; font-size: 11px; }
+.slider-field > span { display: block; margin-bottom: 4px; color: #475569; font-size: 12px; }
+.form-row { display: grid; grid-template-columns: 92px minmax(0, 1fr); align-items: center; gap: 10px; color: #475569; font-size: 12px; }
+.runtime-list { display: grid; gap: 8px; }
+.export-meta { display: grid; gap: 6px; color: #64748b; font-size: 12px; line-height: 1.5; }
+.export-meta span { min-width: 0; display: flex; align-items: center; gap: 5px; word-break: break-all; }
+.export-actions { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 8px; }
+.export-actions :deep(.el-button) { margin-left: 0; }
+.export-preview-video { width: 100%; max-height: 70vh; background: #111827; }
+@media (max-width: 1180px) {
+  .video-demo-layout { grid-template-columns: 1fr; }
+  .video-control-panel { grid-template-columns: repeat(2, minmax(0, 1fr)); }
+}
+@media (max-width: 760px) {
+  .video-stage { min-height: 360px; }
+  .video-control-panel, .video-result-strip { grid-template-columns: 1fr; }
+  .form-row { grid-template-columns: 1fr; }
+}
+</style>

frontend/src/views/workorders/WorkordersView.vue

@@ -15,7 +15,7 @@
     <section v-else class="gis-page-layout workorder-map-layout"><RailwayGisMap :alarms="filteredRows" :selected-id="String(selected?.alarm_id || '')" :layers="{ routes: false, rules: true, alarms: true }" @select="openDetail" /><aside class="gis-detail-panel"><template v-if="selected"><h2>{{ selected.scene }}</h2><p class="entity-id">{{ selected.workorder_id }}</p><el-descriptions :column="1" border size="small"><el-descriptions-item label="状态">{{ statusLabel(selected.status) }}</el-descriptions-item><el-descriptions-item label="责任人">{{ selected.assignee || '待分配' }}</el-descriptions-item><el-descriptions-item label="位置">{{ locationText(selected.location) }}</el-descriptions-item></el-descriptions><div class="drawer-actions"><el-button type="primary" @click="openDetail(selected)">查看工单</el-button></div></template><el-empty v-else description="选择地图点位" /></aside></section>
 
     <el-drawer v-model="detailVisible" title="工单详情" size="660px"><template v-if="selected"><el-steps :active="workorderStep(selected.status)" finish-status="success" align-center><el-step title="工单创建" /><el-step title="接单到场" /><el-step title="现场处置" /><el-step title="复核关闭" /></el-steps><el-descriptions class="workorder-descriptions" :column="1" border><el-descriptions-item label="工单编号"><span class="entity-id">{{ selected.workorder_id }}</span></el-descriptions-item><el-descriptions-item label="关联告警"><el-button link type="primary" @click="router.push(`/alarms/${selected.alarm_id}`)">{{ selected.alarm_id }}</el-button></el-descriptions-item><el-descriptions-item label="隐患场景">{{ selected.scene }} / {{ selected.category }}</el-descriptions-item><el-descriptions-item label="责任人">{{ selected.assignee || '待分配' }}</el-descriptions-item><el-descriptions-item label="处置结果">{{ selected.close_result || '尚未提交' }}</el-descriptions-item><el-descriptions-item label="处置意见">{{ selected.comment || '-' }}</el-descriptions-item><el-descriptions-item label="状态">{{ statusLabel(selected.status) }}</el-descriptions-item></el-descriptions>
-      <div class="drawer-actions workorder-action-bar"><el-button v-if="selected.status === 'created'" type="primary" @click="act(selected, 'ACCEPT')">接收工单</el-button><el-button v-if="selected.status === 'accepted'" type="primary" @click="act(selected, 'DEPART')">出发</el-button><el-button v-if="selected.status === 'en_route'" type="primary" @click="act(selected, 'ARRIVE')">到场签到</el-button><el-button v-if="selected.status === 'on_site'" type="primary" @click="act(selected, 'START_PROCESSING')">开始处置</el-button><el-button v-if="selected.status === 'processing'" type="success" @click="act(selected, 'SUBMIT')">提交处置</el-button><el-button v-if="selected.status === 'returned'" type="primary" @click="act(selected, 'REOPEN')">重新处置</el-button><el-button v-if="['submitted','reviewing'].includes(String(selected.status))" type="success" @click="act(selected, 'APPROVE')">复核通过</el-button><el-button v-if="['submitted','reviewing'].includes(String(selected.status))" type="danger" plain @click="act(selected, 'RETURN')">退回重办</el-button><el-button v-if="selected.status !== 'closed'" @click="redispatch(selected)">人工改派</el-button><el-button v-if="['submitted','reviewing'].includes(String(selected.status))" @click="requestReinspection(selected)">创建复飞任务</el-button><el-button @click="router.push({ path: '/gis', query: { workorderId: selected.workorder_id } })">地图定位</el-button></div>
+      <div class="drawer-actions workorder-action-bar"><el-button v-if="selected.status === 'created'" type="primary" @click="act(selected, 'ACCEPT')">接收工单</el-button><el-button v-if="selected.status === 'accepted'" type="primary" @click="act(selected, 'DEPART')">出发</el-button><el-button v-if="selected.status === 'en_route'" type="primary" @click="act(selected, 'ARRIVE')">到场签到</el-button><el-button v-if="selected.status === 'on_site'" type="primary" @click="act(selected, 'START_PROCESSING')">开始处置</el-button><el-button v-if="selected.status === 'processing'" type="success" @click="act(selected, 'SUBMIT')">提交处置</el-button><el-button v-if="selected.status === 'returned'" type="primary" @click="act(selected, 'REOPEN')">重新处置</el-button><el-button v-if="['submitted','reviewing'].includes(String(selected.status))" type="success" @click="act(selected, 'APPROVE')">复核通过</el-button><el-button v-if="['submitted','reviewing'].includes(String(selected.status))" type="danger" plain @click="act(selected, 'RETURN')">退回重办</el-button><el-button v-if="selected.status !== 'closed'" @click="redispatch(selected)">人工改派</el-button><el-button v-if="['submitted','reviewing'].includes(String(selected.status))" @click="requestReinspection(selected)">创建复飞任务</el-button><el-button @click="openOnMap(selected)">地图定位</el-button></div>
       <div class="subsection-head evidence-head"><div><strong>现场证据</strong><span>整改前、中、后原始文件与摘要</span></div><div><el-select v-model="evidencePhase" size="small"><el-option label="整改前" value="BEFORE" /><el-option label="整改中" value="DURING" /><el-option label="整改后" value="AFTER" /></el-select><el-button type="primary" plain size="small" :icon="Upload" @click="evidenceInput?.click()">上传证据</el-button><input ref="evidenceInput" class="hidden-file-input" type="file" accept="image/*,video/*,audio/*,.pdf" @change="uploadEvidence" /></div></div>
       <div v-if="evidence.length" class="evidence-file-grid"><a v-for="item in evidence" :key="item.attachment_id" :href="workorderEvidenceUrl(String(item.attachment_id))" target="_blank"><el-icon><Picture /></el-icon><span><strong>{{ item.file_name }}</strong><small>{{ evidencePhaseLabel(item.phase) }} · {{ fileSize(item.size_bytes) }}</small></span></a></div><el-empty v-else description="尚未上传现场证据" :image-size="54" />
       <h3 class="drawer-section-title">处置时间轴</h3><el-timeline><el-timeline-item v-for="item in actions" :key="item.action_id" :timestamp="formatDateTime(item.created_at)" placement="top"><strong>{{ actionLabel(item.action_type) }}</strong><p>{{ item.operator_name }} · {{ item.comment || item.result || '状态已更新' }}</p></el-timeline-item></el-timeline>
@@ -44,6 +44,7 @@ function matchTab(row: Row) { if (activeTab.value === "all") return true; if (ac
 function locationText(value: unknown) { const location = safeObject(value); return [location.mileage, location.distance_to_track_m ? `${location.distance_to_track_m}m` : ""].filter(Boolean).join(" / ") || "线路邻近"; }
 function formatDateTime(value: unknown) { return value ? new Date(String(value)).toLocaleString("zh-CN", { hour12: false }) : "-"; }
 function workorderStep(status: unknown) { return ({ created: 1, accepted: 2, en_route: 2, on_site: 2, processing: 3, returned: 3, submitted: 3, reviewing: 3, reinspection_required: 3, closed: 4 } as Record<string, number>)[String(status)] || 1; }
+function openOnMap(row: Row) { router.push({ path: "/gis", query: { workorderId: String(row.workorder_id), alarmId: String(row.alarm_id), taskId: String(row.task_id || "") } }); }
 async function load() { loading.value = true; try { rows.value = await workorders(); const id = String(route.params.workorderId || ""); if (id) { const row = rows.value.find((item) => String(item.workorder_id) === id); if (row) openDetail(row); } } finally { loading.value = false; } }
 async function openDetail(row: Row) { selected.value = row; detailVisible.value = true; router.replace({ path: `/workorders/${row.workorder_id}`, query: route.query }); await loadDetail(row); }
 async function loadDetail(row: Row) { [actions.value, evidence.value] = await Promise.all([workorderActions(String(row.workorder_id)), workorderEvidence(String(row.workorder_id))]); }

frontend/vite.config.ts

@@ -12,6 +12,11 @@ export default defineConfig({
       "/actuator": {
         target: "http://localhost:8080",
         changeOrigin: true
+      },
+      "/vision-api": {
+        target: "http://localhost:8101",
+        changeOrigin: true,
+        rewrite: (path) => path.replace(/^\/vision-api/, "")
       }
     }
   }

infra/docker-compose.yml

@@ -90,6 +90,13 @@ services:
     build:
       context: ../ai-services/vision-inference
     container_name: rail-vision-inference
+    environment:
+      RAIL_RUNTIME_PROFILE: cpu-local
+      RAIL_MODEL_REGISTRY: /app/config/model-registry.json
+      RAIL_MODEL_DIR: /models
+    volumes:
+      - ../runtime/models:/models:ro
+      - ./model-registry/cpu-models.json:/app/config/model-registry.json:ro
     ports:
       - "8101:8101"
     healthcheck:
@@ -102,6 +109,13 @@ services:
     build:
       context: ../ai-services/pointcloud-analysis
     container_name: rail-pointcloud-analysis
+    environment:
+      RAIL_RUNTIME_PROFILE: cpu-local
+      RAIL_MODEL_REGISTRY: /app/config/model-registry.json
+      RAIL_MODEL_DIR: /models
+    volumes:
+      - ../runtime/models:/models:ro
+      - ./model-registry/cpu-models.json:/app/config/model-registry.json:ro
     ports:
       - "8102:8102"
     healthcheck:
@@ -145,7 +159,8 @@ services:
       kafka:
         condition: service_healthy
     environment:
-      SPRING_DATASOURCE_URL: jdbc:postgresql://postgres:5432/rail_inspection
+      SPRING_DATASOURCE_URL: jdbc:postgresql://postgres:5432/rail_inspection?currentSchema=uav_access
+      SPRING_DATASOURCE_SCHEMA: uav_access
       SPRING_DATASOURCE_USERNAME: rail
       SPRING_DATASOURCE_PASSWORD: rail
       SPRING_KAFKA_BOOTSTRAP_SERVERS: kafka:9092

infra/model-registry/cpu-models.json

@@ -19,6 +19,24 @@
       "scale": 0.00392156862745098,
       "color_order": "RGB"
     },
+    {
+      "model_group": "vision-detector",
+      "model_version": "traffic-yolov8n-coco",
+      "active": false,
+      "display_name": "YOLOv8n COCO 交通演示本地 ONNX",
+      "family": "YOLOv8",
+      "runtime": "onnxruntime-cpu",
+      "artifact": "vision-detector/yolov8n-coco/model.onnx",
+      "labels": "vision-detector/yolov8n-coco/labels.txt",
+      "parser": "ultralytics-yolo",
+      "input_size": 640,
+      "batch_size": 1,
+      "precision": "FP32",
+      "mean": [0.0, 0.0, 0.0],
+      "std": [1.0, 1.0, 1.0],
+      "scale": 0.00392156862745098,
+      "color_order": "RGB"
+    },
     {
       "model_group": "vision-segmenter",
       "model_version": "cpu-v1.0.0",
@@ -37,6 +55,24 @@
       "scale": 0.00392156862745098,
       "color_order": "RGB"
     },
+    {
+      "model_group": "vision-segmenter",
+      "model_version": "traffic-yolov8n-seg-coco",
+      "active": false,
+      "display_name": "YOLOv8n-seg COCO 交通实例分割本地 ONNX",
+      "family": "YOLOv8-seg",
+      "runtime": "onnxruntime-cpu",
+      "artifact": "vision-segmenter/yolov8n-seg-coco/model.onnx",
+      "labels": "vision-segmenter/yolov8n-seg-coco/labels.txt",
+      "parser": "ultralytics-yolo-seg",
+      "input_size": 640,
+      "batch_size": 1,
+      "precision": "FP32",
+      "mean": [0.0, 0.0, 0.0],
+      "std": [1.0, 1.0, 1.0],
+      "scale": 0.00392156862745098,
+      "color_order": "RGB"
+    },
     {
       "model_group": "change-detector",
       "model_version": "cpu-v1.0.0",

model-training/.gitignore

@@ -10,4 +10,30 @@ runs/
 artifacts/
 cache/
 .env
-dataset/prepared/
+
+# Local datasets and downloaded experiment bundles
+/dataset/
+/Enhanced-YOLO26s-for-High-Speed-Railway-Foreign-Object-Detection-via-ECA-BiFPN-and-P2-Head-main/
+/*.zip
+
+# Local geospatial inputs
+/dataview/*.tif
+/dataview/*.tiff
+
+# Model checkpoints and exported weights
+*.pt
+*.pth
+*.ckpt
+*.safetensors
+*.onnx
+*.engine
+*.pdparams
+*.pdopt
+*.pdema
+
+# Training/framework caches
+*.cache
+wandb/
+mlruns/
+lightning_logs/
+.ultralytics/

model-training/dataview/DEM1.tif.aux.xml

@@ -0,0 +1,11 @@
+<PAMDataset>
+  <PAMRasterBand band="1">
+    <Metadata>
+      <MDI key="STATISTICS_MINIMUM">1819.5457763672</MDI>
+      <MDI key="STATISTICS_MAXIMUM">1976.1481933594</MDI>
+      <MDI key="STATISTICS_MEAN">1884.4356526837</MDI>
+      <MDI key="STATISTICS_STDDEV">43.989041479334</MDI>
+      <MDI key="STATISTICS_VALID_PERCENT">68.49</MDI>
+    </Metadata>
+  </PAMRasterBand>
+</PAMDataset>

model-training/dataview/dem_final.tif.aux.xml

@@ -0,0 +1,13 @@
+<PAMDataset>
+  <PAMRasterBand band="1">
+    <Description>mean</Description>
+    <Metadata>
+      <MDI key="STATISTICS_APPROXIMATE">YES</MDI>
+      <MDI key="STATISTICS_MINIMUM">1221.9730373912</MDI>
+      <MDI key="STATISTICS_MAXIMUM">1474.4719624217</MDI>
+      <MDI key="STATISTICS_MEAN">1341.5028279639</MDI>
+      <MDI key="STATISTICS_STDDEV">63.088572769604</MDI>
+      <MDI key="STATISTICS_VALID_PERCENT">54.81</MDI>
+    </Metadata>
+  </PAMRasterBand>
+</PAMDataset>

platform/backend/src/main/java/com/ai/trackwalker/capability/api/CapabilityCompletionController.java

@@ -75,6 +75,14 @@ public class CapabilityCompletionController {
         return ApiResponse.ok(service.createRoute(request));
     }
 
+    @PutMapping("/inspection/tasks/{taskId}/route")
+    public ApiResponse<?> bindTaskRoute(
+            @PathVariable String taskId,
+            @Valid @RequestBody CapabilityRequests.BindTaskRouteRequest request
+    ) {
+        return ApiResponse.ok(service.bindTaskRoute(taskId, request));
+    }
+
     @PostMapping("/route-versions/{versionId}/validate")
     public ApiResponse<?> validateRoute(@PathVariable String versionId) {
         return ApiResponse.ok(service.validateRouteVersion(versionId));

platform/backend/src/main/java/com/ai/trackwalker/capability/api/CapabilityRequests.java

@@ -52,6 +52,14 @@ public final class CapabilityRequests {
     ) {
     }
 
+    public record BindTaskRouteRequest(
+            @NotBlank String routeId,
+            String expectedObjectId,
+            Long expectedTaskVersion,
+            String boundBy
+    ) {
+    }
+
     public record CreateConnectionRequest(
             @NotBlank String name,
             @NotBlank String vendorCode,

platform/backend/src/main/java/com/ai/trackwalker/capability/service/CapabilityCompletionService.java

@@ -404,6 +404,90 @@ public class CapabilityCompletionService {
         return result;
     }
 
+    @Transactional
+    public Map<String, Object> bindTaskRoute(
+            String taskId,
+            CapabilityRequests.BindTaskRouteRequest request
+    ) {
+        Map<String, Object> task = requireRow("""
+                select id,status,route_id,object_scope::text as object_scope,version_no
+                from inspection_tasks where id=? for update
+                """, taskId);
+        long taskVersion = ((Number) task.get("version_no")).longValue();
+        if (request.expectedTaskVersion() != null && request.expectedTaskVersion() != taskVersion) {
+            throw conflict("任务状态已变化，请刷新后重新配置航线");
+        }
+        boolean hasMission = Boolean.TRUE.equals(jdbc.queryForObject(
+                "select exists(select 1 from flight_missions where task_id=?)",
+                Boolean.class,
+                taskId
+        ));
+        String blockReason = InspectionTaskWorkflow.routeBindingBlockReason(
+                String.valueOf(task.get("status")),
+                hasMission
+        );
+        if (blockReason != null) {
+            throw conflict(blockReason);
+        }
+
+        Map<String, Object> route = requireRow("""
+                select r.id as route_id,r.name,r.object_id,r.status,
+                       rv.id as route_version_id,rv.status as version_status
+                from routes r
+                left join route_versions rv on rv.id=r.current_version_id
+                where r.id=?
+                """, request.routeId());
+        if (!"PUBLISHED".equals(String.valueOf(route.get("status")))
+                || !"PUBLISHED".equals(String.valueOf(route.get("version_status")))) {
+            throw conflict("只能为任务绑定已发布航线");
+        }
+        String routeObjectId = route.get("object_id") == null
+                ? null : blankToNull(String.valueOf(route.get("object_id")));
+        if (request.expectedObjectId() != null
+                && !request.expectedObjectId().isBlank()
+                && !request.expectedObjectId().equals(routeObjectId)) {
+            throw conflict("所选航线与任务指定巡检对象不一致");
+        }
+        Integer scopedObjectCount = jdbc.queryForObject(
+                "select count(*) from inspection_task_objects where task_id=?",
+                Integer.class,
+                taskId
+        );
+        if (scopedObjectCount != null && scopedObjectCount > 0) {
+            boolean compatible = routeObjectId != null && Boolean.TRUE.equals(jdbc.queryForObject(
+                    "select exists(select 1 from inspection_task_objects where task_id=? and object_id=?)",
+                    Boolean.class,
+                    taskId,
+                    routeObjectId
+            ));
+            if (!compatible) {
+                throw conflict("所选航线不属于任务巡检对象范围");
+            }
+        }
+
+        Instant now = Instant.now();
+        jdbc.update("""
+                update inspection_tasks
+                set route_id=?,updated_at=?,version_no=version_no+1
+                where id=?
+                """, request.routeId(), Timestamp.from(now), taskId);
+        publish("inspection.task.route.bound", Map.of(
+                "task_id", taskId,
+                "route_id", request.routeId(),
+                "route_version_id", String.valueOf(route.get("route_version_id")),
+                "bound_by", defaultString(request.boundBy(), "user-dispatcher")
+        ));
+        Map<String, Object> result = new LinkedHashMap<>();
+        result.put("task_id", taskId);
+        result.put("route_id", request.routeId());
+        result.put("route_name", route.get("name"));
+        result.put("route_version_id", route.get("route_version_id"));
+        result.put("object_id", routeObjectId);
+        result.put("task_version", taskVersion + 1);
+        result.put("status", "BOUND");
+        return result;
+    }
+
     @Transactional
     public Map<String, Object> validateRouteVersion(String versionId) {
         Map<String, Object> version = requireRow(
@@ -846,8 +930,11 @@ public class CapabilityCompletionService {
         String targetStatus = MissionStateMachine.targetForCommand(currentStatus, command);
         int progress = ((Number) mission.get("progress")).intValue();
         if ("SIMULATE_PROGRESS".equals(command)) {
-            if (!"FLYING".equals(currentStatus)) throw conflict("只有飞行中的模拟任务可以推进进度");
+            try {
-            progress = Math.min(100, progress + 25);
+                progress = MissionStateMachine.advanceSimulatedProgress(currentStatus, progress);
+            } catch (IllegalArgumentException exception) {
+                throw conflict(exception.getMessage());
+            }
             if (progress >= 100) targetStatus = "COMPLETED";
         } else if (!MissionStateMachine.canTransition(currentStatus, targetStatus)) {
             throw conflict("飞行任务不能从 " + currentStatus + " 执行 " + command + " 到 " + targetStatus);
@@ -1100,7 +1187,7 @@ public class CapabilityCompletionService {
         List<Map<String, Object>> features = new ArrayList<>();
         for (Map<String, Object> row : jdbc.queryForList("select id,name,object_type,line_id,risk_level,status,ST_AsGeoJSON(geom) as geometry from inspection_objects where status='ACTIVE'")) {
             features.add(feature(row.get("id"), "inspection_object", row.get("geometry"), Map.of(
-                    "name", row.get("name"), "object_type", row.get("object_type"), "line_id", row.get("line_id"),
+                    "object_id", row.get("id"), "name", row.get("name"), "object_type", row.get("object_type"), "line_id", row.get("line_id"),
                     "risk_level", row.get("risk_level"), "status", row.get("status")
             )));
         }
@@ -1230,9 +1317,10 @@ public class CapabilityCompletionService {
                 Ids.next("wo-action"), workorderId, action, from, to, request.operatorId(),
                 defaultString(request.operatorName(), "现场处置人员"), request.result(), request.comment(),
                 Jsonb.write(request.metadata() == null ? Map.of() : request.metadata()), Timestamp.from(now));
-        if ("closed".equals(to)) {
+        jdbc.update("""
-            jdbc.update("update alarms set status='closed', updated_at=? where id=?", Timestamp.from(now), workorder.get("alarm_id"));
+                update alarms set status=?,suppressed=false,suppression_reason=null,updated_at=?
-        }
+                where id=?
+                """, WorkOrderStateMachine.alarmStatusFor(to), Timestamp.from(now), workorder.get("alarm_id"));
         publish("workorder.action.completed", Map.of("workorder_id", workorderId, "action", action, "status", to));
         return Map.of("workorder_id", workorderId, "from_status", from, "status", to, "action", action);
     }

platform/backend/src/main/java/com/ai/trackwalker/capability/uav/InspectionTaskWorkflow.java

@@ -47,10 +47,21 @@ public final class InspectionTaskWorkflow {
         return PRE_FLIGHT_STATUSES.contains(taskStatus) && !hasMission;
     }
 
+    public static String routeBindingBlockReason(String taskStatus, boolean hasMission) {
+        if (!PRE_FLIGHT_STATUSES.contains(taskStatus)) {
+            return "只有待执行任务可以配置航线";
+        }
+        if (hasMission) {
+            return "任务已存在飞行任务，不能更换航线";
+        }
+        return null;
+    }
+
     public static String taskStatusForMission(String missionStatus) {
         return switch (missionStatus) {
             case "FLYING", "PAUSED", "RETURNING" -> "flying";
-            case "COMPLETED", "UPLOADING" -> "data_uploading";
+            case "UPLOADING" -> "data_uploading";
+            case "COMPLETED" -> "completed";
             case "CANCELLED" -> "cancelled";
             case "ABORTED", "FAILED", "DISPATCH_FAILED" -> "failed";
             default -> "dispatched";

platform/backend/src/main/java/com/ai/trackwalker/capability/uav/MissionStateMachine.java

@@ -37,4 +37,13 @@ public final class MissionStateMachine {
             default -> throw new IllegalArgumentException("不支持的飞行任务命令: " + command);
         };
     }
+
+    public static int advanceSimulatedProgress(String currentStatus, int currentProgress) {
+        if (!Set.of("FLYING", "RETURNING").contains(currentStatus)) {
+            throw new IllegalArgumentException("只有飞行中或返航中的模拟任务可以推进进度");
+        }
+        return "RETURNING".equals(currentStatus)
+                ? 100
+                : Math.min(100, Math.max(0, currentProgress) + 25);
+    }
 }

platform/backend/src/main/java/com/ai/trackwalker/capability/workorder/WorkOrderStateMachine.java

@@ -41,4 +41,12 @@ public final class WorkOrderStateMachine {
     public static String target(String action) {
         return TARGETS.get(action);
     }
+
+    public static String alarmStatusFor(String workorderStatus) {
+        return switch (workorderStatus) {
+            case "created", "dispatched" -> "dispatched";
+            case "closed" -> "closed";
+            default -> "processing";
+        };
+    }
 }

platform/backend/src/main/java/com/ai/trackwalker/foundation/service/DependencyFreeCompletionService.java

@@ -240,7 +240,7 @@ public class DependencyFreeCompletionService {
                                 owner_org_id,risk_level,inspection_cycle_days,source_crs,transform_version,
                                 accuracy_m,status,created_at,updated_at
                             )
-                            select ?,?,?,?,?,?,canonical_geometry,?::jsonb,?,?,?,?,?::numeric,'ACTIVE',?,?
+                            select ?,?,?,?,?,?,canonical_geometry,?::jsonb,?,?,?,?,?,?::numeric,'ACTIVE',?,?
                             from inspection_object_import_rows where id=?
                             """,
                     objectId, upper(payload.get("object_type")), payload.get("name"), payload.get("line_id"),
@@ -287,15 +287,15 @@ public class DependencyFreeCompletionService {
                 select count(*) from inspection_task_objects where object_id in (%s)
                 """.formatted(placeholders(objectIds.size())), Integer.class, objectIds.toArray());
         if (referenced > 0) throw conflict("导入对象已被任务引用，不能回滚");
+        jdbc.update("""
+                update inspection_object_import_rows set imported_object_id=null where job_id=?
+                """, jobId);
         if (!objectIds.isEmpty()) {
             jdbc.update("delete from inspection_object_versions where object_id in (%s)".formatted(placeholders(objectIds.size())),
                     objectIds.toArray());
             jdbc.update("delete from inspection_objects where id in (%s)".formatted(placeholders(objectIds.size())),
                     objectIds.toArray());
         }
-        jdbc.update("""
-                update inspection_object_import_rows set imported_object_id=null where job_id=?
-                """, jobId);
         jdbc.update("""
                 update inspection_object_import_jobs
                 set status='ROLLED_BACK',confirmed_by=?,updated_at=?
@@ -331,7 +331,8 @@ public class DependencyFreeCompletionService {
                        schedule_timezone,next_run_at,effective_to
                 from inspection_plans where id=?
                 """, planId);
-        int limit = Math.max(1, Math.min(20, count));
+        boolean manual = "MANUAL".equalsIgnoreCase(String.valueOf(plan.get("trigger_type")));
+        int limit = manual ? 1 : Math.max(1, Math.min(20, count));
         Instant cursor = plan.get("next_run_at") == null
                 ? Instant.now().truncatedTo(ChronoUnit.HOURS)
                 : ((Timestamp) plan.get("next_run_at")).toInstant();
@@ -344,7 +345,7 @@ public class DependencyFreeCompletionService {
                     "scheduled_window_start", start.toString(),
                     "scheduled_window_end", end.toString(),
                     "generation_key", sha256(planId + "|" + start + "|" + end),
-                    "source", "SCHEDULE_PREVIEW"
+                    "source", manual ? "MANUAL_PREVIEW" : "SCHEDULE_PREVIEW"
             ));
             cursor = start;
         }

platform/backend/src/main/java/com/ai/trackwalker/operations/OperationsHealthController.java

@@ -0,0 +1,21 @@
+package com.ai.trackwalker.operations;
+
+import com.ai.trackwalker.api.ApiResponse;
+import org.springframework.web.bind.annotation.GetMapping;
+import org.springframework.web.bind.annotation.RequestMapping;
+import org.springframework.web.bind.annotation.RestController;
+
+@RestController
+@RequestMapping("/api/v1/operations")
+public class OperationsHealthController {
+    private final OperationsHealthService healthService;
+
+    public OperationsHealthController(OperationsHealthService healthService) {
+        this.healthService = healthService;
+    }
+
+    @GetMapping("/health")
+    public ApiResponse<?> health() {
+        return ApiResponse.ok(healthService.snapshot());
+    }
+}

platform/backend/src/main/java/com/ai/trackwalker/operations/OperationsHealthService.java

@@ -0,0 +1,282 @@
+package com.ai.trackwalker.operations;
+
+import com.ai.trackwalker.config.RailProperties;
+import jakarta.annotation.PreDestroy;
+import org.apache.kafka.clients.admin.Admin;
+import org.apache.kafka.clients.admin.AdminClientConfig;
+import org.apache.kafka.clients.admin.DescribeClusterResult;
+import org.apache.kafka.common.Node;
+import org.springframework.boot.web.client.RestTemplateBuilder;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.data.redis.core.RedisCallback;
+import org.springframework.data.redis.core.StringRedisTemplate;
+import org.springframework.http.ResponseEntity;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.kafka.core.KafkaAdmin;
+import org.springframework.stereotype.Service;
+import org.springframework.web.client.RestTemplate;
+
+import java.time.Duration;
+import java.time.Instant;
+import java.util.ArrayList;
+import java.util.Collection;
+import java.util.HashMap;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Locale;
+import java.util.Map;
+import java.util.Objects;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ThreadFactory;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
+
+@Service
+public class OperationsHealthService {
+    private static final Duration HTTP_TIMEOUT = Duration.ofSeconds(3);
+    private static final long KAFKA_TIMEOUT_SECONDS = 3;
+
+    private final List<ProbeDefinition> probes;
+    private final ExecutorService probeExecutor;
+
+    @Autowired
+    public OperationsHealthService(
+            JdbcTemplate jdbc,
+            StringRedisTemplate redis,
+            KafkaAdmin kafkaAdmin,
+            RailProperties properties,
+            RestTemplateBuilder restTemplateBuilder
+    ) {
+        RestTemplate http = restTemplateBuilder
+                .setConnectTimeout(HTTP_TIMEOUT)
+                .setReadTimeout(HTTP_TIMEOUT)
+                .build();
+        this.probes = List.of(
+                definition("platform", "业务平台", "/actuator/health", "服务内自检",
+                        () -> up("运维探测接口已响应")),
+                definition("postgis", "PostGIS", "PostgreSQL / PostGIS", "SQL 探测",
+                        () -> up("PostGIS " + jdbc.queryForObject("select PostGIS_Version()", String.class))),
+                definition("redis", "Redis", redisEndpoint(), "PING",
+                        () -> {
+                            String pong = redis.execute((RedisCallback<String>) connection -> connection.ping());
+                            if (!"PONG".equalsIgnoreCase(pong)) {
+                                throw new IllegalStateException("PING 未返回 PONG");
+                            }
+                            return up("PING=PONG");
+                        }),
+                definition("kafka", "Kafka", kafkaEndpoint(kafkaAdmin), "集群元数据",
+                        () -> kafkaProbe(kafkaAdmin)),
+                definition("minio", "MinIO 对象存储", properties.getStorage().getMinioEndpoint(), "HTTP liveness",
+                        () -> httpProbe(http, properties.getStorage().getMinioEndpoint(), "/minio/health/live")),
+                definition("vision", "视觉推理服务", properties.getAi().getVisionUrl(), "HTTP /health",
+                        () -> httpProbe(http, properties.getAi().getVisionUrl(), "/health")),
+                definition("pointcloud", "点云分析服务", properties.getAi().getPointcloudUrl(), "HTTP /health",
+                        () -> httpProbe(http, properties.getAi().getPointcloudUrl(), "/health")),
+                definition("artifact-installer", "模型安装代理", properties.getArtifactInstaller().getUrl(), "HTTP /health",
+                        () -> httpProbe(http, properties.getArtifactInstaller().getUrl(), "/health")),
+                definition("uav-access", "无人机接入服务", properties.getUavAccess().getUrl(), "Actuator /health",
+                        () -> httpProbe(http, properties.getUavAccess().getUrl(), "/actuator/health"))
+        );
+        this.probeExecutor = newProbeExecutor();
+    }
+
+    OperationsHealthService(List<ProbeDefinition> probes) {
+        this.probes = List.copyOf(probes);
+        this.probeExecutor = newProbeExecutor();
+    }
+
+    public Map<String, Object> snapshot() {
+        Instant checkedAt = Instant.now();
+        List<CompletableFuture<Map<String, Object>>> futures = probes.stream()
+                .map(probe -> CompletableFuture.supplyAsync(() -> runProbe(probe, checkedAt), probeExecutor))
+                .toList();
+        List<Map<String, Object>> services = futures.stream().map(CompletableFuture::join).toList();
+        Map<String, Map<String, Object>> byCode = new HashMap<>();
+        services.forEach(service -> byCode.put(String.valueOf(service.get("code")), service));
+
+        String overall = services.stream().allMatch(service -> "UP".equals(service.get("status")))
+                ? "UP"
+                : "DEGRADED";
+        return Map.of(
+                "status", overall,
+                "checked_at", checkedAt,
+                "services", services,
+                "integrations", integrations(byCode)
+        );
+    }
+
+    @PreDestroy
+    public void shutdown() {
+        probeExecutor.shutdownNow();
+    }
+
+    static ProbeDefinition definition(
+            String code,
+            String name,
+            String endpoint,
+            String check,
+            CheckedProbe probe
+    ) {
+        return new ProbeDefinition(code, name, endpoint, check, probe);
+    }
+
+    static ProbeOutcome up(String detail) {
+        return new ProbeOutcome("UP", detail);
+    }
+
+    static ProbeOutcome down(String detail) {
+        return new ProbeOutcome("DOWN", detail);
+    }
+
+    private Map<String, Object> runProbe(ProbeDefinition definition, Instant checkedAt) {
+        long started = System.nanoTime();
+        ProbeOutcome outcome;
+        try {
+            outcome = Objects.requireNonNull(definition.probe().check(), "探测没有返回结果");
+        } catch (Exception error) {
+            outcome = down(concise(error));
+        }
+        Map<String, Object> row = new LinkedHashMap<>();
+        row.put("code", definition.code());
+        row.put("name", definition.name());
+        row.put("endpoint", definition.endpoint());
+        row.put("check", definition.check());
+        row.put("status", outcome.status());
+        row.put("status_text", "UP".equals(outcome.status()) ? "运行正常" : "连接异常");
+        row.put("detail", outcome.detail());
+        row.put("latency_ms", TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - started));
+        row.put("checked_at", checkedAt);
+        return row;
+    }
+
+    private List<Map<String, Object>> integrations(Map<String, Map<String, Object>> services) {
+        List<Map<String, Object>> rows = new ArrayList<>();
+        rows.add(integration(
+                "uav-task", "无人机任务接口", "SDK/API", "/api/v1/uav/callbacks/mission-status",
+                "航线任务和飞行状态接入", "uav-access", services
+        ));
+        rows.add(integration(
+                "resource-ingest", "多源数据接入", "REST", "/api/v1/inspection/resources/complete",
+                "资源清单与接入完成通知", "platform", services
+        ));
+        rows.add(integration(
+                "workorder-callback", "工单状态回调", "REST", "/api/v1/workorders/callbacks/status",
+                "工单处置状态同步", "platform", services
+        ));
+        rows.add(integration(
+                "event-bus", "无人机事件总线", "Kafka", "uav.access.events.v1",
+                "无人机状态和遥测事件投影", "kafka", services
+        ));
+        return rows;
+    }
+
+    private Map<String, Object> integration(
+            String code,
+            String name,
+            String type,
+            String endpoint,
+            String description,
+            String dependencyCode,
+            Map<String, Map<String, Object>> services
+    ) {
+        Map<String, Object> dependency = services.get(dependencyCode);
+        String dependencyStatus = dependency == null ? "UNKNOWN" : String.valueOf(dependency.get("status"));
+        boolean available = "UP".equals(dependencyStatus);
+        Map<String, Object> row = new LinkedHashMap<>();
+        row.put("code", code);
+        row.put("name", name);
+        row.put("type", type);
+        row.put("endpoint", endpoint);
+        row.put("description", description);
+        row.put("status", available ? "AVAILABLE" : "UNAVAILABLE");
+        row.put("status_text", available
+                ? ("platform".equals(dependencyCode) ? "接口已提供" : "依赖可用")
+                : (dependency == null ? "未探测" : "依赖异常"));
+        row.put("evidence", dependency == null
+                ? "缺少 " + dependencyCode + " 探测结果"
+                : "基于“" + dependency.get("name") + "”实时探测：" + dependency.get("detail"));
+        return row;
+    }
+
+    private static ProbeOutcome httpProbe(RestTemplate http, String baseUrl, String path) {
+        if (baseUrl == null || baseUrl.isBlank()) {
+            return down("服务地址未配置");
+        }
+        String url = baseUrl.endsWith("/")
+                ? baseUrl.substring(0, baseUrl.length() - 1) + path
+                : baseUrl + path;
+        ResponseEntity<Map> response = http.getForEntity(url, Map.class);
+        if (!response.getStatusCode().is2xxSuccessful()) {
+            return down("HTTP " + response.getStatusCode().value());
+        }
+        Map<?, ?> body = response.getBody();
+        Object reportedValue = body == null ? null : body.get("status");
+        String reported = reportedValue == null ? "" : String.valueOf(reportedValue);
+        String normalized = reported.toUpperCase(Locale.ROOT);
+        if (List.of("DOWN", "OUT_OF_SERVICE", "UNAVAILABLE", "FAILED").contains(normalized)) {
+            return down("服务报告状态 " + reported);
+        }
+        return up(reported.isBlank() ? "HTTP " + response.getStatusCode().value() : "服务报告状态 " + reported);
+    }
+
+    private static ProbeOutcome kafkaProbe(KafkaAdmin kafkaAdmin) throws Exception {
+        Map<String, Object> configuration = new HashMap<>(kafkaAdmin.getConfigurationProperties());
+        configuration.put(AdminClientConfig.REQUEST_TIMEOUT_MS_CONFIG, (int) Duration.ofSeconds(KAFKA_TIMEOUT_SECONDS).toMillis());
+        configuration.put(AdminClientConfig.DEFAULT_API_TIMEOUT_MS_CONFIG, (int) Duration.ofSeconds(KAFKA_TIMEOUT_SECONDS).toMillis());
+        Admin admin = Admin.create(configuration);
+        try {
+            DescribeClusterResult cluster = admin.describeCluster();
+            Collection<Node> nodes = cluster.nodes().get(KAFKA_TIMEOUT_SECONDS, TimeUnit.SECONDS);
+            if (nodes.isEmpty()) {
+                return down("集群未返回 broker");
+            }
+            return up("可用 broker " + nodes.size() + " 个");
+        } finally {
+            admin.close(Duration.ZERO);
+        }
+    }
+
+    private static String redisEndpoint() {
+        String host = System.getenv().getOrDefault("SPRING_REDIS_HOST", "localhost");
+        String port = System.getenv().getOrDefault("SPRING_REDIS_PORT", "6379");
+        return host + ":" + port;
+    }
+
+    private static String kafkaEndpoint(KafkaAdmin kafkaAdmin) {
+        Object value = kafkaAdmin.getConfigurationProperties().get(AdminClientConfig.BOOTSTRAP_SERVERS_CONFIG);
+        return value == null ? "未配置" : String.valueOf(value);
+    }
+
+    private static ExecutorService newProbeExecutor() {
+        AtomicInteger sequence = new AtomicInteger();
+        ThreadFactory factory = task -> {
+            Thread thread = new Thread(task, "operations-health-" + sequence.incrementAndGet());
+            thread.setDaemon(true);
+            return thread;
+        };
+        return Executors.newFixedThreadPool(8, factory);
+    }
+
+    private static String concise(Exception error) {
+        Throwable current = error;
+        while (current.getCause() != null && current.getCause() != current) {
+            current = current.getCause();
+        }
+        String message = current.getMessage();
+        String text = current.getClass().getSimpleName() + (message == null || message.isBlank() ? "" : ": " + message);
+        return text.length() > 220 ? text.substring(0, 220) : text;
+    }
+
+    record ProbeDefinition(String code, String name, String endpoint, String check, CheckedProbe probe) {
+    }
+
+    record ProbeOutcome(String status, String detail) {
+    }
+
+    @FunctionalInterface
+    interface CheckedProbe {
+        ProbeOutcome check() throws Exception;
+    }
+}

platform/backend/src/main/java/com/ai/trackwalker/service/PlatformService.java

@@ -2,13 +2,17 @@ package com.ai.trackwalker.service;
 
 import com.ai.trackwalker.api.dto.Requests;
 import com.ai.trackwalker.capability.uav.InspectionTaskWorkflow;
+import com.ai.trackwalker.capability.workorder.WorkOrderStateMachine;
 import com.ai.trackwalker.common.Ids;
 import com.ai.trackwalker.common.Jsonb;
 import org.springframework.jdbc.core.JdbcTemplate;
 import org.springframework.kafka.core.KafkaTemplate;
 import org.springframework.http.HttpStatus;
 import org.springframework.stereotype.Service;
+import org.springframework.transaction.PlatformTransactionManager;
+import org.springframework.transaction.TransactionDefinition;
 import org.springframework.transaction.annotation.Transactional;
+import org.springframework.transaction.support.TransactionTemplate;
 import org.springframework.web.server.ResponseStatusException;
 
 import java.sql.Timestamp;
@@ -23,12 +27,21 @@ public class PlatformService {
     private final AiClient aiClient;
     private final RuleEngine ruleEngine;
     private final KafkaTemplate<String, Object> kafkaTemplate;
+    private final TransactionTemplate transactionTemplate;
 
-    public PlatformService(JdbcTemplate jdbc, AiClient aiClient, RuleEngine ruleEngine, KafkaTemplate<String, Object> kafkaTemplate) {
+    public PlatformService(
+            JdbcTemplate jdbc,
+            AiClient aiClient,
+            RuleEngine ruleEngine,
+            KafkaTemplate<String, Object> kafkaTemplate,
+            PlatformTransactionManager transactionManager
+    ) {
         this.jdbc = jdbc;
         this.aiClient = aiClient;
         this.ruleEngine = ruleEngine;
         this.kafkaTemplate = kafkaTemplate;
+        this.transactionTemplate = new TransactionTemplate(transactionManager);
+        this.transactionTemplate.setPropagationBehavior(TransactionDefinition.PROPAGATION_REQUIRES_NEW);
     }
 
     @Transactional
@@ -150,12 +163,23 @@ public class PlatformService {
                 """);
     }
 
-    @Transactional
     public Map<String, Object> completeResources(Requests.ResourceCompleteRequest request) {
-        List<String> resourceIds = ingestResources(request);
+        Map<String, Object> prepared = transactionTemplate.execute(ignored -> {
-        Map<String, Object> job = createAnalysisJob(new Requests.CreateAnalysisJobRequest(request.taskId(), resourceIds, null, "offline", "normal"));
+            List<String> resourceIds = ingestResources(request);
-        runAnalysis(String.valueOf(job.get("analysis_job_id")));
+            Map<String, Object> job = createAnalysisJob(new Requests.CreateAnalysisJobRequest(
-        return Map.of("accepted", resourceIds.size(), "analysis_job_id", job.get("analysis_job_id"));
+                    request.taskId(), resourceIds, null, "offline", "normal"));
+            return Map.of(
+                    "accepted", resourceIds.size(),
+                    "analysis_job_id", job.get("analysis_job_id")
+            );
+        });
+        if (prepared == null) {
+            throw new IllegalStateException("资源接入事务未返回结果");
+        }
+        Map<String, Object> analysis = runAnalysis(String.valueOf(prepared.get("analysis_job_id")));
+        Map<String, Object> response = new HashMap<>(prepared);
+        response.put("analysis_status", analysis.get("status"));
+        return response;
     }
 
     @Transactional
@@ -224,16 +248,77 @@ public class PlatformService {
                 """);
     }
 
-    @Transactional
     public Map<String, Object> runAnalysis(String jobId) {
-        Map<String, Object> job = jdbc.queryForMap("select *, resource_ids::text as resource_ids_text, scene_set::text as scene_set_text from analysis_jobs where id=?", jobId);
+        AnalysisRunClaim claim = transactionTemplate.execute(ignored -> claimAnalysisRun(jobId));
+        if (claim == null) {
+            throw new IllegalStateException("分析任务状态事务未返回结果");
+        }
+        if (!claim.shouldExecute()) {
+            return currentAnalysisResponse(jobId, claim.job());
+        }
+
+        try {
+            Map<String, Object> result = transactionTemplate.execute(ignored -> executeAnalysis(jobId, claim.job()));
+            if (result == null) {
+                throw new IllegalStateException("分析执行事务未返回结果");
+            }
+            return result;
+        } catch (RuntimeException failure) {
+            try {
+                transactionTemplate.executeWithoutResult(ignored -> markAnalysisFailed(jobId, failure));
+            } catch (RuntimeException persistenceFailure) {
+                failure.addSuppressed(persistenceFailure);
+            }
+            if (failure instanceof ResponseStatusException responseStatusException) {
+                throw responseStatusException;
+            }
+            throw new ResponseStatusException(
+                    HttpStatus.INTERNAL_SERVER_ERROR,
+                    "分析任务执行失败：" + failureMessage(failure),
+                    failure
+            );
+        }
+    }
+
+    private AnalysisRunClaim claimAnalysisRun(String jobId) {
+        List<Map<String, Object>> rows = jdbc.queryForList("""
+                select id, task_id, resource_ids::text as resource_ids_text,
+                       scene_set::text as scene_set_text, status, summary::text as summary_text
+                from analysis_jobs
+                where id=?
+                for update
+                """, jobId);
+        if (rows.isEmpty()) {
+            throw new ResponseStatusException(HttpStatus.NOT_FOUND, "分析任务不存在");
+        }
+        Map<String, Object> job = rows.get(0);
+        String status = String.valueOf(job.get("status"));
+        if ("running".equals(status) || "completed".equals(status)) {
+            return new AnalysisRunClaim(false, job);
+        }
+        if (!"queued".equals(status) && !"failed".equals(status)) {
+            throw new ResponseStatusException(HttpStatus.CONFLICT, "当前分析任务状态不允许执行：" + status);
+        }
+        int updated = jdbc.update("""
+                update analysis_jobs
+                set status='running', summary='{}'::jsonb, completed_at=null
+                where id=? and status=?
+                """, jobId, status);
+        if (updated != 1) {
+            throw new ResponseStatusException(HttpStatus.CONFLICT, "分析任务状态已变化，请刷新后重试");
+        }
+        job.put("status", "running");
+        job.put("summary_text", "{}");
+        return new AnalysisRunClaim(true, job);
+    }
+
+    private Map<String, Object> executeAnalysis(String jobId, Map<String, Object> job) {
         Map<String, Object> task = taskById(String.valueOf(job.get("task_id")));
         List<String> resourceIds = Jsonb.stringList(String.valueOf(job.get("resource_ids_text")));
         List<String> scenes = Jsonb.stringList(String.valueOf(job.get("scene_set_text")));
         int aiResults = 0;
         int alarms = 0;
         int suppressed = 0;
-        jdbc.update("update analysis_jobs set status='running' where id=?", jobId);
         for (String resourceId : resourceIds) {
             Map<String, Object> resource = resourceById(resourceId);
             List<Map<String, Object>> results = aiClient.analyze(resource, scenes);
@@ -251,23 +336,86 @@ public class PlatformService {
             }
         }
         Map<String, Object> summary = Map.of("ai_results", aiResults, "alarms", alarms, "suppressed", suppressed);
-        jdbc.update("update analysis_jobs set status='completed', summary=?::jsonb, completed_at=? where id=?", Jsonb.write(summary), Timestamp.from(Instant.now()), jobId);
+        int completed = jdbc.update("""
+                update analysis_jobs
+                set status='completed', summary=?::jsonb, completed_at=?
+                where id=? and status='running'
+                """, Jsonb.write(summary), Timestamp.from(Instant.now()), jobId);
+        if (completed != 1) {
+            throw new ResponseStatusException(HttpStatus.CONFLICT, "分析任务状态已变化，结果未提交");
+        }
         jdbc.update("update inspection_tasks set status='completed', updated_at=? where id=?", Timestamp.from(Instant.now()), task.get("id"));
         return Map.of("analysis_job_id", jobId, "status", "completed", "summary", summary);
     }
 
-    public List<Map<String, Object>> listAlarms(boolean includeSuppressed) {
+    private Map<String, Object> currentAnalysisResponse(String jobId, Map<String, Object> job) {
-        String sql = "select id as alarm_id, task_id, result_id, scene, category, severity, confidence, location::text as location, evidence::text as evidence, rule_hits::text as rule_hits, status, suppressed, suppression_reason, created_at, updated_at from alarms";
+        Map<String, Object> response = new HashMap<>();
-        if (!includeSuppressed) {
+        response.put("analysis_job_id", jobId);
-            sql += " where suppressed=false";
+        response.put("status", job.get("status"));
+        response.put("summary", Jsonb.map(String.valueOf(job.getOrDefault("summary_text", "{}"))));
+        response.put("idempotent", true);
+        return response;
+    }
+
+    private void markAnalysisFailed(String jobId, RuntimeException failure) {
+        Map<String, Object> summary = Map.of(
+                "error", failureMessage(failure),
+                "failed_at", Instant.now().toString()
+        );
+        jdbc.update("""
+                update analysis_jobs
+                set status='failed', summary=?::jsonb, completed_at=?
+                where id=? and status='running'
+                """, Jsonb.write(summary), Timestamp.from(Instant.now()), jobId);
+    }
+
+    private String failureMessage(Throwable failure) {
+        Throwable cause = failure;
+        while (cause.getCause() != null && cause.getCause() != cause) {
+            cause = cause.getCause();
         }
-        sql += " order by created_at desc";
+        String message = cause.getMessage();
+        if (message == null || message.isBlank()) {
+            message = cause.getClass().getSimpleName();
+        }
+        return message.length() > 1000 ? message.substring(0, 1000) : message;
+    }
+
+    private record AnalysisRunClaim(boolean shouldExecute, Map<String, Object> job) {
+    }
+
+    public List<Map<String, Object>> listAlarms(boolean includeSuppressed) {
+        String sql = """
+                select a.id as alarm_id,a.task_id,a.result_id,a.scene,a.category,a.severity,a.confidence,
+                       a.location::text as location,a.evidence::text as evidence,a.rule_hits::text as rule_hits,
+                       case when a.suppressed then 'suppressed'
+                            when wo.status is null then a.status
+                            when wo.status in ('created','dispatched') then 'dispatched'
+                            when wo.status='closed' then 'closed'
+                            else 'processing' end as status,
+                       a.suppressed,a.suppression_reason,a.created_at,a.updated_at
+                from alarms a
+                left join lateral (
+                    select status from work_orders where alarm_id=a.id order by created_at desc limit 1
+                ) wo on true
+                """;
+        if (!includeSuppressed) {
+            sql += " where a.suppressed=false";
+        }
+        sql += " order by a.created_at desc";
         return jdbc.queryForList(sql);
     }
 
     @Transactional
     public Map<String, Object> decideAlarm(String alarmId, Requests.AlarmDecisionRequest request) {
         String action = request.action() == null ? "confirm" : request.action();
+        List<Map<String, Object>> linkedWorkorders = jdbc.queryForList(
+                "select id,status from work_orders where alarm_id=? order by created_at desc limit 1",
+                alarmId
+        );
+        if ("suppress".equals(action) && !linkedWorkorders.isEmpty()) {
+            throw new ResponseStatusException(HttpStatus.CONFLICT, "告警已生成处置工单，请在工单中完成闭环");
+        }
         String status;
         boolean suppressed;
         String reason = request.reason();
@@ -277,11 +425,15 @@ public class PlatformService {
                 suppressed = true;
             }
             case "reopen" -> {
-                status = "pending";
+                status = linkedWorkorders.isEmpty()
+                        ? "pending"
+                        : WorkOrderStateMachine.alarmStatusFor(String.valueOf(linkedWorkorders.get(0).get("status")));
                 suppressed = false;
             }
             default -> {
-                status = "confirmed";
+                status = linkedWorkorders.isEmpty()
+                        ? "confirmed"
+                        : WorkOrderStateMachine.alarmStatusFor(String.valueOf(linkedWorkorders.get(0).get("status")));
                 suppressed = false;
             }
         }
@@ -326,17 +478,21 @@ public class PlatformService {
                 ? jdbc.queryForMap("select id, alarm_id from work_orders where alarm_id=?", request.alarmId())
                 : jdbc.queryForMap("select id, alarm_id from work_orders where id=?", request.workorderId());
         String workorderId = String.valueOf(workorder.get("id"));
-        jdbc.update("update work_orders set status=?, close_result=?, comment=?, updated_at=? where alarm_id=?",
+        String alarmId = String.valueOf(workorder.get("alarm_id"));
+        jdbc.update("update work_orders set status=?, close_result=?, comment=?, updated_at=? where id=?",
                 request.status() == null ? "closed" : request.status(),
                 request.result(),
                 request.comment(),
                 Timestamp.from(Instant.now()),
-                request.alarmId());
+                workorderId);
-        String alarmStatus = "closed".equals(request.status()) || request.status() == null ? "closed" : "processing";
+        String alarmStatus = WorkOrderStateMachine.alarmStatusFor(
-        jdbc.update("update alarms set status=?, updated_at=? where id=?", alarmStatus, Timestamp.from(Instant.now()), request.alarmId());
+                request.status() == null ? "closed" : request.status()
+        );
+        jdbc.update("update alarms set status=?,suppressed=false,suppression_reason=null,updated_at=? where id=?",
+                alarmStatus, Timestamp.from(Instant.now()), alarmId);
         insertWorkOrderOperation(workorderId, "status_callback", request.operator(), request.result(), request.comment());
-        publish("workorder.closed", Map.of("alarm_id", request.alarmId(), "result", request.result()));
+        publish("workorder.closed", Map.of("alarm_id", alarmId, "result", request.result()));
-        return Map.of("workorder_id", workorderId, "alarm_id", request.alarmId(), "status", request.status() == null ? "closed" : request.status());
+        return Map.of("workorder_id", workorderId, "alarm_id", alarmId, "status", request.status() == null ? "closed" : request.status());
     }
 
     @Transactional
@@ -347,8 +503,8 @@ public class PlatformService {
         Instant now = Instant.now();
         jdbc.update("update work_orders set status=?, close_result=?, comment=?, updated_at=? where id=?",
                 targetStatus, request.result(), request.comment(), Timestamp.from(now), workorderId);
-        jdbc.update("update alarms set status=?, updated_at=? where id=?",
+        jdbc.update("update alarms set status=?,suppressed=false,suppression_reason=null,updated_at=? where id=?",
-                "closed".equals(targetStatus) ? "closed" : "processing", Timestamp.from(now), workorder.get("alarm_id"));
+                WorkOrderStateMachine.alarmStatusFor(targetStatus), Timestamp.from(now), workorder.get("alarm_id"));
         insertWorkOrderOperation(workorderId, "return".equals(action) ? "review_returned" : "review_approved",
                 request.operator(), request.result(), request.comment());
         publish("workorder.reviewed", Map.of("workorder_id", workorderId, "status", targetStatus));
@@ -358,11 +514,21 @@ public class PlatformService {
     @Transactional
     public Map<String, Object> redispatchWorkOrder(String workorderId, String assignee, String reason, String operator) {
         String resolvedAssignee = assignee == null || assignee.isBlank() ? "专业复核人员" : assignee;
+        List<Map<String, Object>> workorders = jdbc.queryForList(
+                "select id,alarm_id from work_orders where id=?",
+                workorderId
+        );
+        if (workorders.isEmpty()) {
+            throw new ResponseStatusException(HttpStatus.NOT_FOUND, "工单不存在");
+        }
+        Map<String, Object> workorder = workorders.get(0);
         int updated = jdbc.update("update work_orders set status='created', assignee=?, comment=?, updated_at=? where id=?",
                 resolvedAssignee, reason, Timestamp.from(Instant.now()), workorderId);
         if (updated == 0) {
             throw new ResponseStatusException(HttpStatus.NOT_FOUND, "工单不存在");
         }
+        jdbc.update("update alarms set status='dispatched',suppressed=false,suppression_reason=null,updated_at=? where id=?",
+                Timestamp.from(Instant.now()), workorder.get("alarm_id"));
         insertWorkOrderOperation(workorderId, "redispatched", operator, "重新派发", reason);
         publish("workorder.redispatched", Map.of("workorder_id", workorderId, "assignee", resolvedAssignee));
         return Map.of("workorder_id", workorderId, "status", "created", "assignee", resolvedAssignee);
@@ -412,7 +578,10 @@ public class PlatformService {
         alarm.put("evidence", Jsonb.map(String.valueOf(alarm.get("evidence"))));
         alarm.put("rule_hits", Jsonb.stringList(String.valueOf(alarm.get("rule_hits"))));
         Map<String, Object> result = resultEvidence(String.valueOf(alarm.get("result_id")));
-        List<Map<String, Object>> relatedWorkorders = jdbc.queryForList("select id as workorder_id, status, assignee, close_result, comment, created_at, updated_at from work_orders where alarm_id=?", alarmId);
+        List<Map<String, Object>> relatedWorkorders = jdbc.queryForList("select id as workorder_id, status, assignee, close_result, comment, created_at, updated_at from work_orders where alarm_id=? order by updated_at desc", alarmId);
+        if (!Boolean.TRUE.equals(alarm.get("suppressed")) && !relatedWorkorders.isEmpty()) {
+            alarm.put("status", WorkOrderStateMachine.alarmStatusFor(String.valueOf(relatedWorkorders.get(0).get("status"))));
+        }
         return Map.of("alarm", alarm, "ai_result", result, "workorders", relatedWorkorders);
     }
 
@@ -541,7 +710,7 @@ public class PlatformService {
                 Jsonb.write(decision.location()),
                 Jsonb.write(Map.of("resource_id", resource.get("id"), "storage_url", resource.get("storage_url"))),
                 Jsonb.write(decision.ruleHits()),
-                decision.suppressed() ? "suppressed" : "pending",
+                decision.suppressed() ? "suppressed" : "dispatched",
                 decision.suppressed(),
                 decision.suppressionReason(),
                 ownerOrgId,

platform/backend/src/main/resources/application.yml

@@ -27,6 +27,7 @@ spring:
       properties:
         spring.json.trusted.packages: "*"
         spring.json.use.type.headers: false
+        spring.json.value.default.type: java.util.LinkedHashMap
     producer:
       key-serializer: org.apache.kafka.common.serialization.StringSerializer
       value-serializer: org.springframework.kafka.support.serializer.JsonSerializer

platform/backend/src/test/java/com/ai/trackwalker/capability/uav/MissionStateMachineTest.java

@@ -36,7 +36,24 @@ class MissionStateMachineTest {
     void taskStatusFollowsMissionLifecycle() {
         assertThat(InspectionTaskWorkflow.taskStatusForMission("DISPATCHED")).isEqualTo("dispatched");
         assertThat(InspectionTaskWorkflow.taskStatusForMission("PAUSED")).isEqualTo("flying");
-        assertThat(InspectionTaskWorkflow.taskStatusForMission("COMPLETED")).isEqualTo("data_uploading");
+        assertThat(InspectionTaskWorkflow.taskStatusForMission("UPLOADING")).isEqualTo("data_uploading");
+        assertThat(InspectionTaskWorkflow.taskStatusForMission("COMPLETED")).isEqualTo("completed");
         assertThat(InspectionTaskWorkflow.taskStatusForMission("ABORTED")).isEqualTo("failed");
     }
+
+    @Test
+    void routeCanOnlyBeBoundBeforeFlightStarts() {
+        assertThat(InspectionTaskWorkflow.routeBindingBlockReason("created", false)).isNull();
+        assertThat(InspectionTaskWorkflow.routeBindingBlockReason("pending", false)).isNull();
+        assertThat(InspectionTaskWorkflow.routeBindingBlockReason("flying", false)).isNotNull();
+        assertThat(InspectionTaskWorkflow.routeBindingBlockReason("created", true)).isNotNull();
+    }
+
+    @Test
+    void returningSimulatorCanAdvanceDirectlyToCompletion() {
+        assertThat(MissionStateMachine.advanceSimulatedProgress("FLYING", 50)).isEqualTo(75);
+        assertThat(MissionStateMachine.advanceSimulatedProgress("RETURNING", 50)).isEqualTo(100);
+        assertThatThrownBy(() -> MissionStateMachine.advanceSimulatedProgress("PAUSED", 50))
+                .isInstanceOf(IllegalArgumentException.class);
+    }
 }

platform/backend/src/test/java/com/ai/trackwalker/capability/workorder/WorkOrderStateMachineTest.java

@@ -22,4 +22,12 @@ class WorkOrderStateMachineTest {
         assertThat(WorkOrderStateMachine.canApply("closed", "RETURN")).isFalse();
         assertThat(WorkOrderStateMachine.supports("UNKNOWN")).isFalse();
     }
+
+    @Test
+    void mapsWorkorderLifecycleBackToAlarmStatus() {
+        assertThat(WorkOrderStateMachine.alarmStatusFor("created")).isEqualTo("dispatched");
+        assertThat(WorkOrderStateMachine.alarmStatusFor("accepted")).isEqualTo("processing");
+        assertThat(WorkOrderStateMachine.alarmStatusFor("submitted")).isEqualTo("processing");
+        assertThat(WorkOrderStateMachine.alarmStatusFor("closed")).isEqualTo("closed");
+    }
 }

platform/backend/src/test/java/com/ai/trackwalker/foundation/service/DependencyFreeCompletionServiceObjectImportTest.java

@@ -0,0 +1,142 @@
+package com.ai.trackwalker.foundation.service;
+
+import com.ai.trackwalker.capability.service.CapabilityCompletionService;
+import org.junit.jupiter.api.Test;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.kafka.core.KafkaTemplate;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.mockito.Mockito.mock;
+
+class DependencyFreeCompletionServiceObjectImportTest {
+
+    @Test
+    void commitUsesOneJdbcArgumentForEveryInspectionObjectPlaceholder() {
+        ObjectImportJdbcTemplate jdbc = new ObjectImportJdbcTemplate();
+        DependencyFreeCompletionService service = service(jdbc);
+
+        Map<String, Object> result = service.commitObjectImport("import-job-123456789012", "reviewer-1");
+
+        assertThat(result).containsEntry("status", "COMMITTED");
+        assertThat((List<?>) result.get("object_ids")).hasSize(1);
+        UpdateCall objectInsert = jdbc.updateCalls.stream()
+                .filter(call -> call.sql().contains("insert into inspection_objects"))
+                .findFirst()
+                .orElseThrow();
+        assertThat(questionMarks(objectInsert.sql())).isEqualTo(16);
+        assertThat(objectInsert.arguments()).hasSize(16);
+        assertThat(objectInsert.arguments()[11]).isEqualTo("DEMO_TRANSFORM_V1");
+        assertThat(objectInsert.arguments()[12]).isEqualTo(5.0);
+        assertThat(jdbc.updateCalls).anyMatch(call -> call.sql().contains("insert into inspection_object_versions"));
+        assertThat(jdbc.updateCalls).anyMatch(call -> call.sql().contains("set status='COMMITTED'"));
+        assertPlaceholderCountsMatch(jdbc);
+    }
+
+    @Test
+    void rollbackRemovesImportedObjectsAndClearsImportLinks() {
+        ObjectImportJdbcTemplate jdbc = new ObjectImportJdbcTemplate();
+        DependencyFreeCompletionService service = service(jdbc);
+
+        Map<String, Object> result = service.rollbackObjectImport("import-job-1", "reviewer-1");
+
+        assertThat(result)
+                .containsEntry("status", "ROLLED_BACK")
+                .containsEntry("rolled_back_objects", 1);
+        assertThat(jdbc.updateCalls).anyMatch(call -> call.sql().contains("delete from inspection_object_versions"));
+        assertThat(jdbc.updateCalls).anyMatch(call -> call.sql().contains("delete from inspection_objects"));
+        assertThat(jdbc.updateCalls).anyMatch(call -> call.sql().contains("set imported_object_id=null"));
+        assertThat(jdbc.updateCalls).anyMatch(call -> call.sql().contains("set status='ROLLED_BACK'"));
+        assertThat(updateIndex(jdbc, "set imported_object_id=null"))
+                .as("import-row foreign keys must be cleared before deleting imported objects")
+                .isLessThan(updateIndex(jdbc, "delete from inspection_objects"));
+        assertPlaceholderCountsMatch(jdbc);
+    }
+
+    private int updateIndex(ObjectImportJdbcTemplate jdbc, String sqlFragment) {
+        for (int index = 0; index < jdbc.updateCalls.size(); index++) {
+            if (jdbc.updateCalls.get(index).sql().contains(sqlFragment)) return index;
+        }
+        return -1;
+    }
+
+    private void assertPlaceholderCountsMatch(ObjectImportJdbcTemplate jdbc) {
+        assertThat(jdbc.updateCalls).allSatisfy(call ->
+                assertThat(call.arguments())
+                        .as("JDBC argument count for %s", call.sql().strip())
+                        .hasSize(questionMarks(call.sql()))
+        );
+    }
+
+    @SuppressWarnings("unchecked")
+    private DependencyFreeCompletionService service(ObjectImportJdbcTemplate jdbc) {
+        return new DependencyFreeCompletionService(
+                jdbc,
+                (KafkaTemplate<String, Object>) mock(KafkaTemplate.class),
+                mock(CapabilityCompletionService.class)
+        );
+    }
+
+    private static int questionMarks(String sql) {
+        return (int) sql.chars().filter(character -> character == '?').count();
+    }
+
+    private record UpdateCall(String sql, Object[] arguments) {
+    }
+
+    private static final class ObjectImportJdbcTemplate extends JdbcTemplate {
+        private final List<UpdateCall> updateCalls = new ArrayList<>();
+
+        @Override
+        public List<Map<String, Object>> queryForList(String sql, Object... args) {
+            if (sql.contains("invalid_count") && sql.contains("from inspection_object_import_jobs")) {
+                return List.of(new HashMap<>(Map.of(
+                        "id", "import-job-123456789012",
+                        "status", "READY",
+                        "invalid_count", 0,
+                        "created_by", "importer-1"
+                )));
+            }
+            if (sql.contains("select id,status from inspection_object_import_jobs")) {
+                return List.of(new HashMap<>(Map.of(
+                        "id", "import-job-1",
+                        "status", "COMMITTED"
+                )));
+            }
+            if (sql.contains("canonical_payload") && sql.contains("from inspection_object_import_rows")) {
+                return List.of(new HashMap<>(Map.of(
+                        "id", "import-row-1",
+                        "canonical_payload", "{\"object_type\":\"BRIDGE\",\"name\":\"一号桥\",\"line_id\":\"line-1\",\"mileage_start\":\"K1\",\"mileage_end\":\"K2\",\"attributes\":{}}"
+                )));
+            }
+            throw new AssertionError("Unexpected query: " + sql);
+        }
+
+        @Override
+        @SuppressWarnings("unchecked")
+        public <T> List<T> queryForList(String sql, Class<T> elementType, Object... args) {
+            if (sql.contains("select imported_object_id")) {
+                return (List<T>) List.of("object-1");
+            }
+            throw new AssertionError("Unexpected typed query: " + sql);
+        }
+
+        @Override
+        public <T> T queryForObject(String sql, Class<T> requiredType, Object... args) {
+            if (sql.contains("from inspection_task_objects")) {
+                return requiredType.cast(0);
+            }
+            throw new AssertionError("Unexpected scalar query: " + sql);
+        }
+
+        @Override
+        public int update(String sql, Object... args) {
+            updateCalls.add(new UpdateCall(sql, args.clone()));
+            return 1;
+        }
+    }
+}

platform/backend/src/test/java/com/ai/trackwalker/foundation/service/DependencyFreeCompletionServicePlanPreviewTest.java

@@ -0,0 +1,108 @@
+package com.ai.trackwalker.foundation.service;
+
+import com.ai.trackwalker.capability.service.CapabilityCompletionService;
+import org.junit.jupiter.api.Test;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.kafka.core.KafkaTemplate;
+
+import java.sql.Timestamp;
+import java.time.Duration;
+import java.time.Instant;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.mockito.Mockito.mock;
+
+class DependencyFreeCompletionServicePlanPreviewTest {
+
+    @Test
+    void manualPlanReturnsOnlyItsSingleExecutionWindow() {
+        Instant scheduledAt = Instant.parse("2026-08-02T01:30:00Z");
+        DependencyFreeCompletionService service = service(plan(
+                "MANUAL",
+                "{\"timing\":\"SCHEDULED\"}",
+                scheduledAt
+        ));
+
+        List<Map<String, Object>> runs = service.previewPlanRuns("plan-manual", 6);
+
+        assertThat(runs).hasSize(1);
+        assertThat(runs.get(0))
+                .containsEntry("sequence", 1)
+                .containsEntry("scheduled_window_start", scheduledAt.toString())
+                .containsEntry("scheduled_window_end", scheduledAt.plus(Duration.ofHours(1)).toString())
+                .containsEntry("source", "MANUAL_PREVIEW");
+        assertThat(String.valueOf(runs.get(0).get("generation_key"))).hasSize(64);
+    }
+
+    @Test
+    void immediateManualPlanUsesTheCurrentSingleWindow() {
+        Instant before = Instant.now().truncatedTo(java.time.temporal.ChronoUnit.HOURS);
+        DependencyFreeCompletionService service = service(plan(
+                "MANUAL",
+                "{\"timing\":\"NOW\"}",
+                null
+        ));
+
+        List<Map<String, Object>> runs = service.previewPlanRuns("plan-manual-now", 20);
+        Instant after = Instant.now().truncatedTo(java.time.temporal.ChronoUnit.HOURS);
+
+        assertThat(runs).hasSize(1);
+        assertThat(Instant.parse(String.valueOf(runs.get(0).get("scheduled_window_start"))))
+                .isIn(before, after);
+        assertThat(runs.get(0)).containsEntry("source", "MANUAL_PREVIEW");
+    }
+
+    @Test
+    void periodicPlanStillRespectsRequestedPreviewCount() {
+        DependencyFreeCompletionService service = service(plan(
+                "PERIODIC",
+                "{\"frequency\":\"DAILY\",\"interval_days\":1,\"execution_time\":\"09:00\"}",
+                Instant.parse("2026-08-02T01:00:00Z")
+        ));
+
+        List<Map<String, Object>> runs = service.previewPlanRuns("plan-periodic", 3);
+
+        assertThat(runs).hasSize(3);
+        assertThat(runs).allSatisfy(run -> assertThat(run).containsEntry("source", "SCHEDULE_PREVIEW"));
+    }
+
+    @SuppressWarnings("unchecked")
+    private DependencyFreeCompletionService service(Map<String, Object> plan) {
+        return new DependencyFreeCompletionService(
+                new PlanJdbcTemplate(plan),
+                (KafkaTemplate<String, Object>) mock(KafkaTemplate.class),
+                mock(CapabilityCompletionService.class)
+        );
+    }
+
+    private Map<String, Object> plan(String triggerType, String scheduleConfig, Instant nextRunAt) {
+        Map<String, Object> plan = new HashMap<>();
+        plan.put("id", "plan-1");
+        plan.put("trigger_type", triggerType);
+        plan.put("schedule_rule", "PERIODIC".equals(triggerType) ? "DAILY" : null);
+        plan.put("schedule_config", scheduleConfig);
+        plan.put("schedule_timezone", "Asia/Shanghai");
+        plan.put("next_run_at", nextRunAt == null ? null : Timestamp.from(nextRunAt));
+        plan.put("effective_to", null);
+        return plan;
+    }
+
+    private static final class PlanJdbcTemplate extends JdbcTemplate {
+        private final Map<String, Object> plan;
+
+        private PlanJdbcTemplate(Map<String, Object> plan) {
+            this.plan = plan;
+        }
+
+        @Override
+        public List<Map<String, Object>> queryForList(String sql, Object... args) {
+            if (sql.contains("from inspection_plans")) {
+                return List.of(new HashMap<>(plan));
+            }
+            throw new AssertionError("Unexpected query: " + sql);
+        }
+    }
+}

platform/backend/src/test/java/com/ai/trackwalker/operations/OperationsHealthServiceTest.java

@@ -0,0 +1,55 @@
+package com.ai.trackwalker.operations;
+
+import org.junit.jupiter.api.Test;
+
+import java.util.List;
+import java.util.Map;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+class OperationsHealthServiceTest {
+
+    @Test
+    void keepsProbeResultsIndependentAndMarksAffectedIntegrations() {
+        OperationsHealthService service = new OperationsHealthService(List.of(
+                OperationsHealthService.definition(
+                        "platform", "业务平台", "/actuator/health", "服务内自检",
+                        () -> OperationsHealthService.up("接口已响应")
+                ),
+                OperationsHealthService.definition(
+                        "kafka", "Kafka", "kafka:9092", "集群元数据",
+                        () -> {
+                            throw new IllegalStateException("broker unavailable");
+                        }
+                ),
+                OperationsHealthService.definition(
+                        "uav-access", "无人机接入服务", "uav-access:8091", "HTTP /health",
+                        () -> OperationsHealthService.down("HTTP 503")
+                )
+        ));
+
+        try {
+            Map<String, Object> snapshot = service.snapshot();
+
+            assertThat(snapshot).containsEntry("status", "DEGRADED");
+            assertThat(rows(snapshot, "services"))
+                    .extracting(row -> row.get("code") + ":" + row.get("status"))
+                    .containsExactly("platform:UP", "kafka:DOWN", "uav-access:DOWN");
+            assertThat(rows(snapshot, "integrations"))
+                    .filteredOn(row -> List.of("resource-ingest", "event-bus", "uav-task").contains(row.get("code")))
+                    .extracting(row -> row.get("code") + ":" + row.get("status_text"))
+                    .containsExactlyInAnyOrder(
+                            "resource-ingest:接口已提供",
+                            "event-bus:依赖异常",
+                            "uav-task:依赖异常"
+                    );
+        } finally {
+            service.shutdown();
+        }
+    }
+
+    @SuppressWarnings("unchecked")
+    private List<Map<String, Object>> rows(Map<String, Object> snapshot, String key) {
+        return (List<Map<String, Object>>) snapshot.get(key);
+    }
+}

platform/backend/src/test/java/com/ai/trackwalker/service/PlatformServiceAnalysisTest.java

@@ -0,0 +1,169 @@
+package com.ai.trackwalker.service;
+
+import org.junit.jupiter.api.Test;
+import org.springframework.jdbc.core.JdbcTemplate;
+import org.springframework.kafka.core.KafkaTemplate;
+import org.springframework.transaction.PlatformTransactionManager;
+import org.springframework.transaction.TransactionDefinition;
+import org.springframework.transaction.TransactionStatus;
+import org.springframework.transaction.support.SimpleTransactionStatus;
+import org.springframework.web.server.ResponseStatusException;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+import static org.assertj.core.api.Assertions.assertThat;
+import static org.assertj.core.api.Assertions.assertThatThrownBy;
+import static org.mockito.ArgumentMatchers.anyList;
+import static org.mockito.ArgumentMatchers.anyMap;
+import static org.mockito.Mockito.mock;
+import static org.mockito.Mockito.never;
+import static org.mockito.Mockito.times;
+import static org.mockito.Mockito.verify;
+import static org.mockito.Mockito.when;
+
+class PlatformServiceAnalysisTest {
+
+    @Test
+    void completedJobIsIdempotentAndDoesNotCreateDuplicateResults() {
+        RecordingJdbcTemplate jdbc = new RecordingJdbcTemplate("queued");
+        AiClient aiClient = mock(AiClient.class);
+        when(aiClient.analyze(anyMap(), anyList())).thenReturn(List.of(new HashMap<>(Map.of(
+                "scene", "塔吊",
+                "category", "塔吊",
+                "confidence", 0.91,
+                "measurements", Map.of("distance_to_track_m", 83.2)
+        ))));
+        PlatformService service = service(jdbc, aiClient);
+
+        Map<String, Object> first = service.runAnalysis("job-1");
+        Map<String, Object> second = service.runAnalysis("job-1");
+
+        assertThat(first.get("status")).isEqualTo("completed");
+        assertThat(second).containsEntry("status", "completed").containsEntry("idempotent", true);
+        verify(aiClient, times(1)).analyze(anyMap(), anyList());
+        assertThat(jdbc.updates.stream().filter(sql -> sql.contains("insert into ai_results"))).hasSize(1);
+        assertThat(jdbc.updates.stream().filter(sql -> sql.contains("insert into alarms"))).hasSize(1);
+        assertThat(jdbc.updates.stream().filter(sql -> sql.contains("insert into work_orders"))).hasSize(1);
+    }
+
+    @Test
+    void runningJobReturnsCurrentStateWithoutStartingAnotherExecution() {
+        RecordingJdbcTemplate jdbc = new RecordingJdbcTemplate("running");
+        AiClient aiClient = mock(AiClient.class);
+        PlatformService service = service(jdbc, aiClient);
+
+        Map<String, Object> result = service.runAnalysis("job-1");
+
+        assertThat(result).containsEntry("status", "running").containsEntry("idempotent", true);
+        verify(aiClient, never()).analyze(anyMap(), anyList());
+        assertThat(jdbc.updates).isEmpty();
+    }
+
+    @Test
+    void failedRetryPersistsFailedStatusAfterInferenceError() {
+        RecordingJdbcTemplate jdbc = new RecordingJdbcTemplate("failed");
+        AiClient aiClient = mock(AiClient.class);
+        when(aiClient.analyze(anyMap(), anyList())).thenThrow(new IllegalStateException("inference unavailable"));
+        PlatformService service = service(jdbc, aiClient);
+
+        assertThatThrownBy(() -> service.runAnalysis("job-1"))
+                .isInstanceOf(ResponseStatusException.class)
+                .hasMessageContaining("分析任务执行失败")
+                .hasMessageContaining("inference unavailable");
+
+        assertThat(jdbc.updates).anyMatch(sql -> sql.contains("set status='running'"));
+        assertThat(jdbc.updates).anyMatch(sql -> sql.contains("set status='failed'"));
+        assertThat(jdbc.jobStatus).isEqualTo("failed");
+        assertThat(jdbc.summary).contains("inference unavailable");
+    }
+
+    @SuppressWarnings("unchecked")
+    private PlatformService service(RecordingJdbcTemplate jdbc, AiClient aiClient) {
+        return new PlatformService(
+                jdbc,
+                aiClient,
+                new RuleEngine(),
+                (KafkaTemplate<String, Object>) mock(KafkaTemplate.class),
+                new NoOpTransactionManager()
+        );
+    }
+
+    private static final class RecordingJdbcTemplate extends JdbcTemplate {
+        private final List<String> updates = new ArrayList<>();
+        private String jobStatus;
+        private String summary = "{}";
+
+        private RecordingJdbcTemplate(String jobStatus) {
+            this.jobStatus = jobStatus;
+        }
+
+        @Override
+        public List<Map<String, Object>> queryForList(String sql, Object... args) {
+            if (sql.contains("from analysis_jobs")) {
+                Map<String, Object> job = new HashMap<>();
+                job.put("id", "job-1");
+                job.put("task_id", "task-1");
+                job.put("resource_ids_text", "[\"resource-1\"]");
+                job.put("scene_set_text", "[\"塔吊\"]");
+                job.put("status", jobStatus);
+                job.put("summary_text", summary);
+                return List.of(job);
+            }
+            return List.of();
+        }
+
+        @Override
+        public Map<String, Object> queryForMap(String sql, Object... args) {
+            if (sql.contains("from inspection_tasks")) {
+                return new HashMap<>(Map.of(
+                        "id", "task-1",
+                        "line_id", "line-1",
+                        "scene_set_text", "[\"塔吊\"]"
+                ));
+            }
+            if (sql.contains("from inspection_resources")) {
+                return new HashMap<>(Map.of(
+                        "id", "resource-1",
+                        "task_id", "task-1",
+                        "storage_url", "s3://bucket/resource-1.jpg",
+                        "metadata", "{}"
+                ));
+            }
+            throw new AssertionError("Unexpected query: " + sql);
+        }
+
+        @Override
+        public int update(String sql, Object... args) {
+            updates.add(sql);
+            if (sql.contains("update analysis_jobs") && sql.contains("set status='running'")) {
+                jobStatus = "running";
+                summary = "{}";
+            } else if (sql.contains("update analysis_jobs") && sql.contains("set status='completed'")) {
+                jobStatus = "completed";
+                summary = String.valueOf(args[0]);
+            } else if (sql.contains("update analysis_jobs") && sql.contains("set status='failed'")) {
+                jobStatus = "failed";
+                summary = String.valueOf(args[0]);
+            }
+            return 1;
+        }
+    }
+
+    private static final class NoOpTransactionManager implements PlatformTransactionManager {
+        @Override
+        public TransactionStatus getTransaction(TransactionDefinition definition) {
+            return new SimpleTransactionStatus();
+        }
+
+        @Override
+        public void commit(TransactionStatus status) {
+        }
+
+        @Override
+        public void rollback(TransactionStatus status) {
+        }
+    }
+}

scripts/capability_completion_test.ps1

@@ -1,6 +1,6 @@
 $ErrorActionPreference = "Stop"
 
-$baseUrl = if ($env:RAIL_WEB_BASE_URL) { $env:RAIL_WEB_BASE_URL.TrimEnd("/") } else { "http://localhost:8088" }
+$baseUrl = if ($env:RAIL_WEB_BASE_URL) { $env:RAIL_WEB_BASE_URL.TrimEnd("/") } else { "http://127.0.0.1:8088" }
 $apiUrl = "$baseUrl/api/v1"
 $runKey = (Get-Date).ToUniversalTime().ToString("yyyyMMddHHmmssfff")
 $tempPng = Join-Path ([IO.Path]::GetTempPath()) "rail-capability-$runKey.png"
@@ -77,6 +77,19 @@ function Invoke-WorkOrderAction {
     }
 }
 
+function Invoke-WorkflowAction {
+    param([string]$BusinessType, [string]$BusinessId, [string]$Action, [int]$Sequence)
+    $isApproval = $Action -in @("APPROVE", "REJECT")
+    return Invoke-RailApi -Method POST -Path "/workflows/$BusinessType/$BusinessId/actions" -Body @{
+        action = $Action
+        operator_id = if ($isApproval) { "user-approver" } else { "user-dispatcher" }
+        operator_name = if ($isApproval) { "System Approver" } else { "System Dispatcher" }
+        opinion = "capability completion regression"
+        attachments = @()
+        idempotency_key = "$runKey-workflow-$BusinessType-$BusinessId-$Sequence"
+    }
+}
+
 try {
     Write-Host "Checking capability registry and spatial objects..."
     $completion = Invoke-RailApi -Method GET -Path "/capabilities/completion"
@@ -101,6 +114,9 @@ try {
     }
     $validation = Invoke-RailApi -Method POST -Path "/route-versions/$($route.route_version_id)/validate"
     if (-not $validation.valid) { throw "Route validation failed" }
+    Invoke-WorkflowAction -BusinessType "ROUTE_VERSION" -BusinessId $route.route_version_id -Action "SUBMIT" -Sequence 1 | Out-Null
+    $routeApproval = Invoke-WorkflowAction -BusinessType "ROUTE_VERSION" -BusinessId $route.route_version_id -Action "APPROVE" -Sequence 2
+    if ($routeApproval.state -ne "APPROVED") { throw "Route approval failed" }
     $published = Invoke-RailApi -Method POST -Path "/route-versions/$($route.route_version_id)/publish?approvedBy=system-tester"
     if ($published.status -ne "PUBLISHED") { throw "Route publication failed" }
 
@@ -108,7 +124,8 @@ try {
     $plan = Invoke-RailApi -Method POST -Path "/inspection/plans" -Body @{
         name = "Capability regression plan $runKey"
         plan_type = "SPECIAL"
-        schedule_rule = "MANUAL"
+        trigger_type = "MANUAL"
+        schedule_config = @{ timing = "NOW" }
         object_ids = @($object.object_id)
         scene_set = @("FOREIGN_OBJECT", "FENCE_DAMAGE")
         priority = "high"
@@ -120,6 +137,9 @@ try {
     $generated = Invoke-RailApi -Method POST -Path "/inspection/plans/$($plan.plan_id)/generate-tasks"
     $taskId = @($generated.task_ids)[0]
     if (-not $taskId) { throw "Plan did not generate an inspection task" }
+    Invoke-WorkflowAction -BusinessType "TASK" -BusinessId $taskId -Action "SUBMIT" -Sequence 3 | Out-Null
+    $taskApproval = Invoke-WorkflowAction -BusinessType "TASK" -BusinessId $taskId -Action "APPROVE" -Sequence 4
+    if ($taskApproval.state -ne "APPROVED") { throw "Task approval failed" }
 
     Write-Host "Checking simulator connection and executing a flight mission..."
     $connections = Invoke-RailApi -Method GET -Path "/uav/connections"

scripts/navigation_pages_test.ps1

@@ -1,6 +1,6 @@
 $ErrorActionPreference = "Stop"
 
-$baseUrl = if ($env:RAIL_WEB_BASE_URL) { $env:RAIL_WEB_BASE_URL.TrimEnd("/") } else { "http://localhost:8088" }
+$baseUrl = if ($env:RAIL_WEB_BASE_URL) { $env:RAIL_WEB_BASE_URL.TrimEnd("/") } else { "http://127.0.0.1:8088" }
 $apiUrl = "$baseUrl/api/v1"
 
 Write-Host "Checking navigation routes..."
@@ -87,7 +87,35 @@ $missionBody = @{
 $mission = Invoke-RestMethod -Method Post -Uri "$apiUrl/uav/callbacks/mission-status" -ContentType "application/json" -Body $missionBody -TimeoutSec 30
 if ($mission.data.status -ne "flying") { throw "Mission status update failed" }
 
-$cancel = Invoke-RestMethod -Method Post -Uri "$apiUrl/inspection/tasks/$taskId/cancel" -ContentType "application/json" -Body "{}" -TimeoutSec 30
+$completedMissionBody = @{
+    task_id = $taskId
+    vendor = "DJI"
+    vendor_mission_id = "mission-navigation-test"
+    status = "completed"
+    uav_id = "UAV-TEST"
+    timestamp = (Get-Date).ToUniversalTime().ToString("o")
+    position = @{ longitude = 116.1; latitude = 39.1 }
+} | ConvertTo-Json -Depth 4
+$completedMission = Invoke-RestMethod -Method Post -Uri "$apiUrl/uav/callbacks/mission-status" -ContentType "application/json" -Body $completedMissionBody -TimeoutSec 30
+if ($completedMission.data.status -ne "completed") { throw "Mission completion update failed" }
+
+# Cancellation is only valid before a mission has been created. Use a separate
+# pending task so this test covers both lifecycle paths without violating the
+# production safety guard.
+$cancelTaskBody = @{
+    external_task_id = "NAVIGATION-CANCEL-TEST"
+    line_id = "line-test"
+    mileage_start = "K12+000"
+    mileage_end = "K13+000"
+    route_id = "route-navigation-cancel-test"
+    scene_set = @("浜哄憳鍏ヤ镜")
+    priority = "normal"
+    planned_start_time = (Get-Date).ToUniversalTime().AddMinutes(15).ToString("o")
+} | ConvertTo-Json
+$cancelTask = Invoke-RestMethod -Method Post -Uri "$apiUrl/inspection/tasks" -ContentType "application/json" -Body $cancelTaskBody -TimeoutSec 30
+$cancelTaskId = $cancelTask.data.task_id
+if (-not $cancelTaskId) { throw "Cancellation task creation did not return task_id" }
+$cancel = Invoke-RestMethod -Method Post -Uri "$apiUrl/inspection/tasks/$cancelTaskId/cancel" -ContentType "application/json" -Body "{}" -TimeoutSec 30
 if ($cancel.data.status -ne "cancelled") { throw "Task cancellation failed" }
 
 Write-Host "Checking alarm decision operation..."
@@ -96,15 +124,21 @@ $alarmId = @($alarmsResponse.data.alarms | Where-Object { $_.status -ne "closed"
 if ($alarmId) {
     $confirmBody = @{ action = "confirm"; reason = "页面操作回归"; operator = "测试人员" } | ConvertTo-Json
     $confirmed = Invoke-RestMethod -Method Post -Uri "$apiUrl/alarms/$alarmId/decision" -ContentType "application/json" -Body $confirmBody -TimeoutSec 30
-    if ($confirmed.data.status -ne "confirmed") { throw "Alarm confirmation failed" }
+    $validConfirmationStates = @("confirmed", "dispatched", "processing", "closed")
+    if ($validConfirmationStates -notcontains $confirmed.data.status) {
+        throw "Alarm confirmation failed: unexpected status $($confirmed.data.status)"
+    }
 }
 
 $result = [ordered]@{
     route_count = $routes.Count
     endpoint_count = $endpointChecks.Count
     created_task_id = $taskId
-    task_final_status = $cancel.data.status
+    task_final_status = $completedMission.data.status
+    cancelled_task_id = $cancelTaskId
+    cancelled_task_status = $cancel.data.status
     alarm_decision_checked = [bool]$alarmId
+    alarm_decision_status = if ($alarmId) { $confirmed.data.status } else { $null }
     endpoints = $summary
 }
 $result | ConvertTo-Json -Depth 5

scripts/verify_cpu_runtime.ps1

@@ -20,8 +20,8 @@ if (-not $spatial.open3d_available) {
 $models = @($vision.models) + @($spatial.models)
 $groups = @($models | Select-Object -ExpandProperty model_group -Unique)
 $expected = @("vision-detector", "vision-segmenter", "change-detector", "pointcloud-analyzer", "thermal-analyzer")
-if ($models.Count -ne 5) {
+if ($groups.Count -ne 5) {
-    throw "CPU runtime must expose exactly five model groups, received $($models.Count)."
+    throw "CPU runtime must expose exactly five model groups, received $($groups.Count)."
 }
 foreach ($group in $expected) {
     if ($group -notin $groups) {

uav-access-service/src/main/resources/application.yml

@@ -7,9 +7,11 @@ spring:
   jackson:
     property-naming-strategy: SNAKE_CASE
   datasource:
-    url: ${SPRING_DATASOURCE_URL:jdbc:postgresql://localhost:5432/rail_inspection}
+    url: ${SPRING_DATASOURCE_URL:jdbc:postgresql://localhost:5432/rail_inspection?currentSchema=uav_access}
     username: ${SPRING_DATASOURCE_USERNAME:rail}
     password: ${SPRING_DATASOURCE_PASSWORD:rail}
+    hikari:
+      schema: ${SPRING_DATASOURCE_SCHEMA:uav_access}
   flyway:
     enabled: true
     create-schemas: true

visualization-demo/.gitignore

@@ -0,0 +1,2 @@
+outputs/*
+!outputs/.gitkeep

visualization-demo/outputs/.gitkeep

@@ -0,0 +1 @@
+

visualization-demo/video-realtime-ai-demo-design.md

@@ -0,0 +1,778 @@
+# 视频实时目标检测与图像分割可视化演示设计文档
+
+## 1. 文档信息
+
+| 项目 | 内容 |
+| --- | --- |
+| 文档版本 | V0.2 |
+| 编制日期 | 2026-08-01 |
+| 目标目录 | `visualization-demo/` |
+| 目标页面 | 上传本地视频并在播放过程中按需启用目标检测、图像分割实时叠加，并可生成完整结果视频保存到本地 |
+| 适用视频 | 无人机航拍视频、手机拍摄视频、铁路巡检样例视频 |
+| 默认部署形态 | 前端 Vue 3 + Vite，后端 FastAPI + OpenCV + ONNX Runtime GPU |
+
+## 2. 目标与边界
+
+### 2.1 建设目标
+
+1. 在项目根目录新增一个独立的可视化演示目录，用于沉淀视频 AI 演示方案、后续页面和服务适配文件。
+2. 页面支持上传本地视频，使用浏览器原生视频播放器播放，不强制把完整视频上传到后端。
+3. 页面提供目标检测开关和图像分割开关，用户打开后对当前播放帧进行实时推理。
+4. 推理结果以叠加层方式显示在视频画面上：目标检测显示框、类别、置信度；图像分割显示半透明掩膜或轮廓。
+5. 运行时优先使用本机 RTX 3090 的 GPU 推理能力；模型制品缺失时必须给出明确的降级状态，而不是伪装成真实推理。
+6. 页面支持生成完整检测/分割结果视频：把模型检测框、类别、置信度和分割掩膜渲染进整段视频，保存到本地输出目录。
+7. 完整结果视频同时保存结构化结果 JSON、运行参数、模型版本和日志，便于后续复核、演示和样本回流。
+
+### 2.2 不在首版范围内
+
+1. 不自动对所有上传视频做全量分析；只有用户点击“生成结果视频”时才上传完整视频并启动离线任务。
+2. 不在 Git 仓库中提交大模型权重、视频样例和大体积推理制品。
+3. 不在首版引入 SAM 一类提示式大模型作为默认分割方案。它适合交互式精分割，但不适合首版实时视频开关。
+4. 不把演示结果直接标记为生产告警结果。演示页面只展示当前模型能力和性能状态。
+
+## 3. 当前项目依据
+
+### 3.1 已发现的工程基础
+
+| 模块 | 当前状态 | 对本设计的影响 |
+| --- | --- | --- |
+| `frontend/` | Vue 3 + Vite + Element Plus，已有路由、侧边栏和 API 服务封装 | 可新增一个演示视图并接入现有导航 |
+| `frontend/vite.config.ts` | `/api` 代理到 `http://localhost:8080` | 首版可继续走统一后端代理，也可临时直连 `vision-inference` |
+| `ai-services/vision-inference/` | FastAPI，已有 `/api/v1/inference/detect`、模型加载、运行时状态接口 | 可复用现有模型注册、ONNX Runtime、结果解析能力 |
+| `ai-services/vision-inference/requirements-server.txt` | 包含 `onnxruntime-gpu`、OpenCV、FastAPI | 适合本机 RTX 3090 推理服务 |
+| `infra/model-registry/cpu-models.json` | 已登记 `vision-detector`、`vision-segmenter` CPU 模型位 | 可作为 CPU 降级和开发默认配置 |
+| `infra/model-registry/server-models.json` | 已登记 GPU 侧 `vision-detector`、`vision-segmenter`、`thermal-analyzer` 等模型位 | 可作为 GPU 推理默认配置 |
+| `runtime/models/` | 已有模型目录约定，说明生产铁路权重不会自动下载 | 后续模型制品放这里或其子目录，不提交 Git |
+
+### 3.2 本机硬件结论
+
+已通过 `nvidia-smi` 确认本机 GPU：
+
+```text
+NVIDIA GeForce RTX 3090, 24576 MiB, Driver 566.36
+```
+
+因此首选方案为 `ONNX Runtime GPU + CUDAExecutionProvider`。RTX 3090 的 24GB 显存足够支撑轻量检测模型和实时语义/实例分割模型在演示页面中并行切换，但仍需要对输入分辨率、推理帧率和并发请求做节流。
+
+## 4. 用户体验设计
+
+### 4.1 首屏布局
+
+页面首屏即为可操作演示台，不做营销式落地页。
+
+```text
+┌──────────────────────────────────────────────────────────────┐
+│ 顶部：页面标题、运行状态、GPU/CPU Provider、模型加载状态       │
+├──────────────────────────────────────────────┬───────────────┤
+│                                              │ 上传视频       │
+│ 视频播放区                                   │ 检测开关       │
+│ - video 元素                                 │ 分割开关       │
+│ - detection canvas                           │ 阈值/帧率      │
+│ - segmentation canvas                        │ 模型状态       │
+│                                              │ 延迟/FPS       │
+│                                              │ 生成结果视频   │
+│                                              │ 本地输出路径   │
+├──────────────────────────────────────────────┴───────────────┤
+│ 底部：当前帧结果列表、类别筛选、时间轴命中点、错误/降级提示     │
+└──────────────────────────────────────────────────────────────┘
+```
+
+### 4.2 核心交互
+
+1. 用户点击上传区域选择视频文件。
+2. 前端用 `URL.createObjectURL(file)` 生成本地播放地址，视频原文件不默认上传。
+3. 用户播放视频。
+4. 用户打开目标检测开关后，前端按设定 FPS 从视频抽帧，发送当前帧到后端检测接口。
+5. 用户打开图像分割开关后，同一抽帧流程增加分割任务。
+6. 后端返回归一化坐标结果，前端根据当前视频显示尺寸绘制叠加层。
+7. 如果视频暂停，默认暂停抽帧推理；保留最后一帧叠加结果。
+8. 如果拖动进度条，清空旧结果并从新时间点继续推理。
+9. 用户点击“生成结果视频”后，前端上传完整视频和当前模型参数，后端启动离线任务。
+10. 后端逐帧或按配置抽帧推理、插值/复用结果并把检测框与分割掩膜绘制进输出视频。
+11. 任务完成后页面展示本地保存路径、视频预览地址、结果 JSON 和处理耗时。
+
+### 4.3 页面控件
+
+| 控件 | 类型 | 说明 |
+| --- | --- | --- |
+| 上传视频 | 文件选择按钮 | 支持 `mp4`、`webm`、浏览器可解码的 `mov` |
+| 目标检测 | 开关 | 控制 `vision-detector` 或通用检测模型 |
+| 图像分割 | 开关 | 控制 `vision-segmenter` 或通用实例分割模型 |
+| 检测置信度 | 滑块 | 默认 0.45，范围 0.05 到 0.95 |
+| 分割阈值 | 滑块 | 默认 0.5，范围 0.1 到 0.9 |
+| 推理帧率 | 步进器或滑块 | GPU 默认检测 8 FPS、分割 3 FPS；CPU 自动降低 |
+| 最大推理宽度 | 下拉 | 默认 960，选项 640、960、1280 |
+| 结果显示 | 勾选项 | 类别标签、置信度、轮廓、掩膜透明度 |
+| 模型预热 | 按钮 | 主动加载模型，减少第一次打开开关的延迟 |
+| 生成结果视频 | 按钮 | 上传完整视频，后台生成带检测/分割叠加的本地结果视频 |
+| 输出目录 | 只读文本/打开按钮 | 显示 `visualization-demo/outputs/video-runs/<run_id>/` |
+| 任务进度 | 进度条 | 展示已处理帧数、总帧数、预计剩余时间 |
+| 结果下载 | 按钮 | 下载生成后的 `annotated.mp4`、`results.json` 和 `run-metadata.json` |
+
+## 5. 技术架构
+
+### 5.1 总体架构
+
+```mermaid
+flowchart LR
+    A["浏览器视频文件"] --> B["HTMLVideoElement 播放"]
+    B --> C["Canvas 抽帧"]
+    C --> D["FrameScheduler 节流与丢帧"]
+    D --> E["POST /api/v1/video-demo/infer-frame"]
+    E --> F["FastAPI 视频演示接口"]
+    F --> G["OpenCV 解码单帧"]
+    G --> H["ONNX Runtime GPU"]
+    H --> I1["vision-detector"]
+    H --> I2["vision-segmenter"]
+    I1 --> J["统一归一化结果"]
+    I2 --> J
+    J --> K["前端 OverlayRenderer"]
+    K --> L["检测框/分割掩膜叠加"]
+```
+
+### 5.1.1 完整结果视频生成链路
+
+```mermaid
+flowchart LR
+    A["用户点击生成结果视频"] --> B["上传完整视频与模型参数"]
+    B --> C["POST /api/v1/video-demo/export-jobs"]
+    C --> D["本地任务目录 video-runs/<run_id>"]
+    D --> E["OpenCV VideoCapture 逐帧读取"]
+    E --> F["按配置缩放推理帧"]
+    F --> G["检测/分割 ONNX 推理"]
+    G --> H["OverlayComposer 绘制框和掩膜"]
+    H --> I["OpenCV VideoWriter 输出 annotated.mp4"]
+    G --> J["保存 per-frame results.jsonl"]
+    I --> K["GET /api/v1/video-demo/export-jobs/<run_id>"]
+    J --> K
+    K --> L["页面展示保存路径和下载入口"]
+```
+
+### 5.2 前端模块
+
+建议新增：
+
+```text
+frontend/src/views/visualization-demo/VideoAiDemoView.vue
+frontend/src/components/video-ai/VideoStage.vue
+frontend/src/components/video-ai/VideoAiControlPanel.vue
+frontend/src/components/video-ai/ResultTimeline.vue
+frontend/src/composables/useVideoFrameInference.ts
+frontend/src/services/videoDemoApi.ts
+```
+
+职责划分：
+
+| 模块 | 职责 |
+| --- | --- |
+| `VideoAiDemoView.vue` | 页面容器，维护视频文件、开关、阈值和模型状态 |
+| `VideoStage.vue` | 视频播放、叠加层尺寸同步、拖动/暂停事件 |
+| `VideoAiControlPanel.vue` | 上传、开关、阈值、FPS、模型预热 |
+| `ResultTimeline.vue` | 展示当前视频时间附近的检测/分割结果 |
+| `useVideoFrameInference.ts` | 抽帧、请求调度、请求取消、丢帧策略 |
+| `videoDemoApi.ts` | 封装能力查询、帧推理、模型预热接口 |
+
+### 5.3 后端模块
+
+建议在 `ai-services/vision-inference/app/` 新增：
+
+```text
+video_demo_routes.py
+video_demo_schemas.py
+frame_runtime.py
+video_export_runtime.py
+```
+
+职责划分：
+
+| 模块 | 职责 |
+| --- | --- |
+| `video_demo_routes.py` | 暴露能力查询、模型预热、单帧推理接口 |
+| `video_demo_schemas.py` | 定义请求参数和响应结构 |
+| `frame_runtime.py` | 复用现有 ONNX Runtime，对上传帧进行检测/分割 |
+| `video_export_runtime.py` | 负责完整视频读取、推理、叠加渲染、本地文件写入和进度记录 |
+
+首版使用 HTTP 单帧接口，避免 WebSocket 复杂度。若后续需要更高帧率，再增加 WebSocket 或本地共享内存方案。
+
+完整结果视频生成使用后台任务，不阻塞实时预览接口。首版可使用 FastAPI `BackgroundTasks` 或进程内任务队列；后续再接入 Celery、Redis Queue 或平台现有任务调度。
+
+## 6. 接口设计
+
+### 6.1 查询运行能力
+
+```http
+GET /api/v1/video-demo/capabilities
+```
+
+响应示例：
+
+```json
+{
+  "runtime": "onnxruntime-gpu",
+  "execution_provider": "CUDAExecutionProvider",
+  "gpu": {
+    "name": "NVIDIA GeForce RTX 3090",
+    "memory_total_mb": 24576
+  },
+  "models": [
+    {
+      "model_group": "vision-detector",
+      "model_version": "v1.0.0",
+      "display_name": "PP-YOLOE-SOD-s",
+      "artifact_installed": true,
+      "loaded": false
+    },
+    {
+      "model_group": "vision-segmenter",
+      "model_version": "v1.0.0",
+      "display_name": "PP-LiteSeg-STDC1",
+      "artifact_installed": true,
+      "loaded": false
+    }
+  ],
+  "recommended": {
+    "max_inference_width": 960,
+    "detection_fps": 8,
+    "segmentation_fps": 3
+  }
+}
+```
+
+### 6.2 模型预热
+
+```http
+POST /api/v1/video-demo/warmup
+Content-Type: application/json
+```
+
+请求：
+
+```json
+{
+  "models": ["vision-detector", "vision-segmenter"]
+}
+```
+
+### 6.3 单帧推理
+
+```http
+POST /api/v1/video-demo/infer-frame
+Content-Type: multipart/form-data
+```
+
+字段：
+
+| 字段 | 类型 | 说明 |
+| --- | --- | --- |
+| `frame` | file | 当前视频帧，建议 JPEG |
+| `session_id` | string | 前端生成的本次视频会话 ID |
+| `timestamp_ms` | number | 视频当前时间 |
+| `source_width` | number | 视频原始宽度 |
+| `source_height` | number | 视频原始高度 |
+| `detect_enabled` | boolean | 是否执行检测 |
+| `segment_enabled` | boolean | 是否执行分割 |
+| `confidence_threshold` | number | 检测置信度阈值 |
+| `mask_threshold` | number | 分割阈值 |
+| `max_detections` | number | 最大检测数量 |
+
+响应：
+
+```json
+{
+  "session_id": "demo-20260801-001",
+  "timestamp_ms": 12345,
+  "frame_id": "demo-20260801-001:12345",
+  "source": {
+    "width": 1920,
+    "height": 1080
+  },
+  "runtime": {
+    "provider": "CUDAExecutionProvider",
+    "total_latency_ms": 42.7,
+    "decode_latency_ms": 3.1,
+    "detection_latency_ms": 18.4,
+    "segmentation_latency_ms": 21.2
+  },
+  "results": {
+    "detections": [
+      {
+        "category": "person",
+        "confidence": 0.91,
+        "bbox": [0.42, 0.18, 0.53, 0.71],
+        "model_group": "vision-detector",
+        "model_version": "v1.0.0"
+      }
+    ],
+    "segments": [
+      {
+        "category": "roadbed",
+        "confidence": 1.0,
+        "polygon": [[0.1, 0.7], [0.5, 0.45], [0.92, 0.74], [0.1, 0.7]],
+        "area_ratio": 0.18,
+        "model_group": "vision-segmenter",
+        "model_version": "v1.0.0"
+      }
+    ]
+  },
+  "warnings": []
+}
+```
+
+### 6.4 错误与降级响应
+
+如果模型制品未安装：
+
+```json
+{
+  "results": {
+    "detections": [],
+    "segments": []
+  },
+  "warnings": [
+    {
+      "code": "MODEL_ARTIFACT_MISSING",
+      "message": "vision-detector 模型制品未安装，当前未执行真实推理"
+    }
+  ]
+}
+```
+
+页面必须用明显状态展示该警告。
+
+### 6.5 创建完整结果视频任务
+
+```http
+POST /api/v1/video-demo/export-jobs
+Content-Type: multipart/form-data
+```
+
+字段：
+
+| 字段 | 类型 | 说明 |
+| --- | --- | --- |
+| `video` | file | 完整原始视频文件 |
+| `detect_enabled` | boolean | 是否写入检测结果 |
+| `segment_enabled` | boolean | 是否写入分割结果 |
+| `confidence_threshold` | number | 检测置信度阈值 |
+| `mask_threshold` | number | 分割阈值 |
+| `max_inference_width` | number | 推理输入最大宽度 |
+| `output_fps_policy` | string | `source` 保持源视频 FPS，`fixed` 使用指定输出 FPS |
+| `analysis_stride` | number | 每隔多少帧执行一次模型推理，默认 1 |
+| `reuse_last_result` | boolean | 非推理帧是否复用上一帧结果，默认 true |
+
+响应：
+
+```json
+{
+  "run_id": "video-run-20260801-030512",
+  "status": "queued",
+  "output_dir": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512",
+  "status_url": "/api/v1/video-demo/export-jobs/video-run-20260801-030512"
+}
+```
+
+### 6.6 查询完整结果视频任务
+
+```http
+GET /api/v1/video-demo/export-jobs/{run_id}
+```
+
+响应：
+
+```json
+{
+  "run_id": "video-run-20260801-030512",
+  "status": "running",
+  "progress": {
+    "processed_frames": 420,
+    "total_frames": 2400,
+    "percent": 17.5,
+    "elapsed_seconds": 34.2,
+    "eta_seconds": 161.3
+  },
+  "outputs": {
+    "annotated_video": null,
+    "results_json": null,
+    "metadata_json": null,
+    "log": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512/run.log"
+  },
+  "warnings": []
+}
+```
+
+完成后：
+
+```json
+{
+  "run_id": "video-run-20260801-030512",
+  "status": "succeeded",
+  "progress": {
+    "processed_frames": 2400,
+    "total_frames": 2400,
+    "percent": 100
+  },
+  "outputs": {
+    "annotated_video": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512/annotated.mp4",
+    "results_json": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512/results.json",
+    "results_jsonl": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512/results.jsonl",
+    "metadata_json": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512/run-metadata.json",
+    "log": "D:/workspace/AItrackwalker/visualization-demo/outputs/video-runs/video-run-20260801-030512/run.log"
+  },
+  "warnings": []
+}
+```
+
+### 6.7 下载或预览本地输出文件
+
+```http
+GET /api/v1/video-demo/export-jobs/{run_id}/files/{file_name}
+```
+
+允许的 `file_name` 首版固定为：
+
+1. `annotated.mp4`
+2. `results.json`
+3. `results.jsonl`
+4. `run-metadata.json`
+5. `run.log`
+
+接口必须校验 `run_id` 和文件名，禁止路径穿越。
+
+## 7. 模型选型
+
+### 7.1 首选模型
+
+| 功能 | 首选模型 | 原因 | 运行方式 |
+| --- | --- | --- | --- |
+| 目标检测 | `vision-detector`，PP-YOLOE-SOD-s ONNX | 项目已有模型注册、PaddleDetection 输出解析和 RailFOD23 训练基线，适合无人机铁路小目标 | ONNX Runtime GPU，输入 640 到 1280 |
+| 图像分割 | `vision-segmenter`，PP-LiteSeg-STDC1 ONNX | 轻量语义分割模型，适合实时分割路基、边坡、积水、裂缝等区域 | ONNX Runtime GPU，输入 512 到 1024 |
+
+说明：
+
+1. `vision-detector` 当前项目文档中已有 RailFOD23 PP-YOLOE-SOD-s 训练记录，但生产可用仍依赖 ONNX 导出、一致性验证和模型制品安装。
+2. `vision-segmenter` 已有注册位和解析器，真实效果依赖铁路现场或目标场景分割数据微调后的模型制品。
+3. 页面不能把基线回退结果当成真实模型结果。响应中必须保留 `execution_mode`、`artifact_installed`、`fallback_reason`。
+
+### 7.2 通用演示备用模型
+
+如果首版需要覆盖手机随拍视频中的人员、车辆、常见物体，并且项目铁路模型制品尚未就绪，可以增加一组独立的演示模型注册位：
+
+| 功能 | 备用模型 | 用途 | 说明 |
+| --- | --- | --- | --- |
+| 通用检测 | YOLOv8n/YOLOv8s ONNX 或同级轻量 COCO 检测模型 | 手机视频、通用航拍视频的人员和车辆检测 | 只作为演示模型，不替代铁路业务模型 |
+| 通用分割 | YOLOv8n-seg/YOLOv8s-seg ONNX 或同级轻量实例分割模型 | 通用物体实例掩膜叠加 | 比 SAM 更适合实时视频开关 |
+
+备用模型建议放在：
+
+```text
+runtime/models/demo/
+├── object-detector/
+│   ├── model.onnx
+│   ├── labels.txt
+│   └── model-card.md
+└── instance-segmenter/
+    ├── model.onnx
+    ├── labels.txt
+    └── model-card.md
+```
+
+### 7.3 为什么首版不选 SAM
+
+SAM 类模型更适合用户点击、框选或文本提示后的高质量交互分割。当前需求是视频播放过程中打开开关即实时处理，核心指标是端到端延迟、稳定帧率和叠加可读性。轻量语义分割或实例分割模型更贴合首版目标。
+
+## 8. 实时性能策略
+
+### 8.1 前端节流
+
+1. 同一功能只允许一个未完成请求在飞行中。
+2. 如果后端尚未返回，下一帧直接丢弃，不排队。
+3. 检测和分割使用独立帧率：检测默认更高，分割默认更低。
+4. 视频暂停、隐藏标签页或拖动进度条时停止抽帧。
+5. 每次请求带 `timestamp_ms`，前端只渲染与当前播放时间接近的结果。
+
+### 8.2 输入缩放
+
+1. 视频显示可以保持原始清晰度，推理帧默认缩放到最大宽度 960。
+2. 无人机 4K 视频不直接以原始 4K 输入推理。
+3. 检测小目标场景可允许切换到 1280，但 UI 要提示会增加延迟。
+4. 分割默认 512 或 1024，优先保证稳定帧率。
+
+### 8.3 后端并发
+
+1. `vision-inference` 继续使用模型懒加载和最大并发控制。
+2. GPU 模式建议单进程、单 GPU、有限并发，避免多个请求同时抢占显存。
+3. 模型预热后保留最近使用模型，避免频繁加载/卸载。
+4. 检测和分割同时开启时，首版可顺序推理；后续再做并行或 TensorRT 优化。
+
+### 8.4 目标性能
+
+| 模式 | 默认输入 | 目标体验 |
+| --- | --- | --- |
+| RTX 3090，检测 | 960 宽 | 6 到 10 FPS，端到端延迟小于 120ms |
+| RTX 3090，分割 | 768 或 960 宽 | 2 到 5 FPS，端到端延迟小于 250ms |
+| RTX 3090，检测+分割 | 960 宽 | 检测保持较高频，分割低频刷新 |
+| CPU 降级 | 640 宽 | 1 到 2 FPS，仅用于功能验证 |
+
+### 8.5 完整视频生成策略
+
+1. 完整视频任务可以比实时预览慢，但必须有可见进度。
+2. 默认 `analysis_stride=1`，逐帧推理并写入完整结果；当视频过长或分割开启后耗时过高时，可选择每 2 到 5 帧推理一次并复用最近结果。
+3. 输出视频默认保持源视频分辨率和 FPS；推理帧可缩放，绘制结果再映射回源视频坐标。
+4. 使用 `mp4v` 或部署环境可用的 H.264 编码输出 `annotated.mp4`。如果 H.264 编码器不可用，降级为 `mp4v` 并在 `run-metadata.json` 记录编码器。
+5. 每处理 N 帧写入一次进度文件，服务重启后仍能读取已有任务状态。
+6. 输出目录按任务隔离，避免多次导出互相覆盖。
+
+## 9. 前端渲染细节
+
+### 9.1 叠加层坐标
+
+后端统一返回归一化坐标，前端根据视频当前渲染尺寸转换：
+
+```ts
+const x = normalizedX * renderedVideoWidth;
+const y = normalizedY * renderedVideoHeight;
+```
+
+如果视频使用 `object-fit: contain`，需要计算黑边偏移：
+
+```text
+drawX = videoOffsetX + normalizedX * actualVideoDrawWidth
+drawY = videoOffsetY + normalizedY * actualVideoDrawHeight
+```
+
+### 9.2 叠加层分层
+
+```text
+Video 元素
+Segmentation Canvas  半透明掩膜
+Detection Canvas     检测框和标签
+Interaction Layer    hover、选中、结果高亮
+```
+
+分割掩膜透明度默认 0.35，检测框和标签始终在最上层，避免被掩膜遮挡。
+
+### 9.3 结果生命周期
+
+1. 每个结果绑定 `timestamp_ms`。
+2. 当前播放时间距离结果超过 500ms 时隐藏或淡出。
+3. 拖动进度条后清空所有未确认结果。
+4. 视频文件更换后清空会话、请求、模型状态以外的页面状态。
+
+## 10. 后端实现细节
+
+### 10.1 复用现有运行时
+
+现有 `CpuVisionRuntime` 已具备：
+
+1. 模型注册表读取。
+2. ONNX Runtime Session 懒加载。
+3. OpenCV 图像解码。
+4. 检测框解析和语义分割 polygon 解析。
+5. 模型缺失时的受控回退。
+
+新增单帧接口时建议优先复用这些能力，避免另起一套推理引擎。
+
+### 10.2 建议新增方法
+
+在运行时层增加数组输入方法，减少把帧写入临时文件的开销：
+
+```python
+def infer_image_array(
+    self,
+    image: np.ndarray,
+    model_group: str,
+    parameters: dict[str, Any],
+    model_version: str | None = None,
+) -> RuntimeOutcome:
+    ...
+```
+
+HTTP 接口负责把上传的 JPEG/PNG 解码为 `np.ndarray`，然后调用该方法。
+
+### 10.3 结果统一
+
+当前 `InferenceResult.geometry` 支持 `BBox` 和 `Polygon`。视频演示接口可以将其转换为前端更直接的结构：
+
+| 后端 geometry | 前端字段 |
+| --- | --- |
+| `{"type": "BBox", "coordinates": [x1,y1,x2,y2]}` | `bbox` |
+| `{"type": "Polygon", "coordinates": [[[x,y],...]]}` | `polygon` |
+
+### 10.4 完整结果视频本地存储
+
+输出根目录：
+
+```text
+visualization-demo/outputs/video-runs/
+└── <run_id>/
+    ├── source.mp4
+    ├── annotated.mp4
+    ├── results.json
+    ├── results.jsonl
+    ├── run-metadata.json
+    └── run.log
+```
+
+文件说明：
+
+| 文件 | 说明 |
+| --- | --- |
+| `source.mp4` | 上传后的原始视频副本；如用户不希望保存原始视频，可在任务完成后自动删除 |
+| `annotated.mp4` | 带检测框、类别、置信度、分割掩膜和时间戳的完整结果视频 |
+| `results.json` | 汇总结果，包含视频信息、模型信息、总体统计和逐帧结果索引 |
+| `results.jsonl` | 每行一帧或一个推理时间点，便于增量写入和中断恢复 |
+| `run-metadata.json` | 运行参数、模型版本、阈值、输入缩放、Provider、开始/结束时间 |
+| `run.log` | 任务日志、警告、异常和性能统计 |
+
+建议默认保存原始视频副本，便于结果复核；同时在页面提供“任务完成后删除原始视频”的选项。输出目录应加入 Git 忽略规则，避免把视频和结果文件提交进仓库。
+
+### 10.5 视频叠加渲染规则
+
+1. 检测框颜色按类别稳定分配；标签显示类别、置信度和目标编号。
+2. 分割掩膜使用半透明填充，边界使用同类高亮描边。
+3. 检测框绘制在分割掩膜之上。
+4. 左上角可绘制时间戳、模型版本、Provider 和当前帧号。
+5. 右下角可绘制“Demo / 非生产告警”水印，避免演示视频被误用为生产判定。
+6. 输出视频中的中文字体优先使用系统字体；如果 OpenCV 环境不支持中文绘制，则标签使用英文类别名或 PIL 字体渲染后合成。
+
+## 11. 安全与资源约束
+
+1. 实时预览默认不上传完整视频，只上传抽取后的单帧。
+2. 只有用户点击“生成结果视频”时才上传完整视频，并在页面展示本地保存目录。
+3. 单帧上传大小限制建议 2MB。
+4. 页面接受的视频文件大小建议限制为 2GB，并在前端提示浏览器解码能力限制。
+5. 后端拒绝非图像帧、超大帧和异常 MIME。
+6. 完整视频任务必须限制输出根目录，禁止用户指定任意本地写入路径。
+7. 输出视频、原始视频副本和 JSON 结果不进入 Git，使用 `visualization-demo/outputs/` 本地保存。
+8. 模型文件不进入 Git，使用 `runtime/models/` 或部署挂载目录。
+9. 浏览器页面销毁时释放 `ObjectURL`，避免内存泄漏。
+
+## 12. 实施计划
+
+### 阶段 1：文档与目录
+
+1. 新建 `visualization-demo/`。
+2. 输出本设计文档。
+
+### 阶段 2：前端静态演示台
+
+1. 新增 `/visualization-demo` 路由。
+2. 实现视频上传、播放、暂停、进度条适配。
+3. 实现检测和分割开关，但先只展示空结果和运行状态。
+4. 加入 canvas 叠加层尺寸同步。
+
+### 阶段 3：后端单帧推理接口
+
+1. 新增 `GET /api/v1/video-demo/capabilities`。
+2. 新增 `POST /api/v1/video-demo/warmup`。
+3. 新增 `POST /api/v1/video-demo/infer-frame`。
+4. 复用现有模型注册表、ONNX Runtime 和解析器。
+5. 对模型缺失、Provider 不可用、输入错误输出明确警告。
+
+### 阶段 4：前后端联调
+
+1. 前端抽帧并发送 JPEG。
+2. 后端返回检测框和分割 polygon。
+3. 前端按视频显示尺寸叠加结果。
+4. 实现丢帧、取消、拖动进度清理和延迟统计。
+
+### 阶段 5：模型制品与 GPU 验证
+
+1. 安装或挂载 `vision-detector` ONNX 制品。
+2. 安装或挂载 `vision-segmenter` ONNX 制品。
+3. 设置 `RAIL_MODEL_REGISTRY=infra/model-registry/server-models.json`。
+4. 设置 `RAIL_MODEL_DIR=runtime/models` 或实际模型挂载目录。
+5. 设置 `RAIL_EXECUTION_PROVIDER=CUDAExecutionProvider`。
+6. 验证 `/api/v1/runtime` 中 `execution_provider_ready=true`。
+
+### 阶段 6：体验打磨
+
+1. 加入模型状态条、降级提示、实时 FPS 和延迟。
+2. 加入类别筛选、结果 hover 高亮。
+3. 加入截图导出或当前帧结果 JSON 导出。
+4. 加入页面级错误恢复。
+
+### 阶段 7：完整结果视频本地导出
+
+1. 新增 `POST /api/v1/video-demo/export-jobs`。
+2. 新增 `GET /api/v1/video-demo/export-jobs/{run_id}`。
+3. 新增 `GET /api/v1/video-demo/export-jobs/{run_id}/files/{file_name}`。
+4. 实现 `video_export_runtime.py`，用 OpenCV 读取完整视频、调用模型推理、绘制叠加层并写入 `annotated.mp4`。
+5. 保存 `results.json`、`results.jsonl`、`run-metadata.json` 和 `run.log`。
+6. 前端增加任务进度、取消入口、完成后本地路径展示和下载按钮。
+7. 为 `visualization-demo/outputs/` 增加 Git 忽略规则。
+
+## 13. 验收标准
+
+| 类别 | 标准 |
+| --- | --- |
+| 视频播放 | 上传 `mp4` 后可播放、暂停、拖动、重新上传 |
+| 检测开关 | 打开后出现真实检测请求；关闭后停止请求并清除或保留最后结果可配置 |
+| 分割开关 | 打开后出现真实分割请求；掩膜与视频画面位置一致 |
+| 双开关 | 检测和分割可同时开启，页面不卡死，请求不无限排队 |
+| 模型状态 | 模型缺失、CPU 降级、GPU Provider 不可用都有明确提示 |
+| 性能 | RTX 3090 下 960 宽输入检测体验达到 6 FPS 以上 |
+| 坐标准确 | 改变窗口大小、视频比例或全屏时叠加层仍对齐 |
+| 资源释放 | 换视频、离开页面后无明显内存持续增长 |
+| 结果视频 | 点击生成后可在本地目录得到 `annotated.mp4`，视频中包含检测框和分割掩膜 |
+| 结果文件 | 同一任务目录包含 `results.json`、`results.jsonl`、`run-metadata.json` 和 `run.log` |
+| 任务进度 | 长视频处理时页面可查看进度、耗时、预计剩余时间和失败原因 |
+
+## 14. 测试计划
+
+### 14.1 前端测试
+
+1. 抽帧调度：请求未返回时不排队。
+2. 视频尺寸变化：不同宽高比视频下 overlay 对齐。
+3. 开关状态：快速开关不会留下后台请求循环。
+4. 错误展示：后端 500、模型缺失、网络断开均有状态提示。
+
+### 14.2 后端测试
+
+1. `capabilities` 在模型存在和不存在时均返回可解释状态。
+2. `infer-frame` 接收 JPEG/PNG 并拒绝非法文件。
+3. 检测结果 bbox 保持 0 到 1 归一化。
+4. 分割 polygon 点位保持 0 到 1 归一化。
+5. GPU Provider 不可用时降级到 CPU 或返回明确错误。
+6. `export-jobs` 能生成本地任务目录并拒绝非法文件名访问。
+7. 完整视频生成中断或失败时保留 `run.log` 和错误状态。
+
+### 14.3 联调测试
+
+1. 无人机航拍视频：检查小目标框是否稳定。
+2. 手机人物/车辆视频：如果启用通用备用模型，检查常见类别是否可见。
+3. 4K 视频：验证缩放推理和叠加对齐。
+4. 长视频：连续播放 10 分钟观察内存、显存和接口延迟。
+5. 完整结果视频：对 30 秒、3 分钟、10 分钟样例分别生成 `annotated.mp4`，检查音视频时长、画面同步和输出文件完整性。
+
+## 15. 主要风险与对策
+
+| 风险 | 影响 | 对策 |
+| --- | --- | --- |
+| 项目铁路模型 ONNX 制品尚未就绪 | 无法展示真实铁路检测/分割 | 页面显示模型缺失；可接入通用演示模型临时验证交互 |
+| 手机视频和铁路模型类别不匹配 | 检测不到普通物体 | 使用通用备用模型作为演示 profile |
+| 4K 视频抽帧过大 | 延迟高、显存压力大 | 前端缩放到 640/960/1280 后上传 |
+| 检测和分割同时开启导致卡顿 | 页面体验下降 | 独立 FPS、单请求在飞、分割低频刷新 |
+| 视频编码浏览器不支持 | 无法播放部分 `mov` | 页面提示转换为 H.264 MP4 或 WebM |
+| 坐标叠加错位 | 演示可信度下降 | 统一归一化坐标并处理 `object-fit` 黑边 |
+| 完整视频导出耗时较长 | 用户以为任务卡死 | 后台任务、进度文件、预计剩余时间和日志 |
+| 输出视频体积过大 | 占用本地磁盘 | 输出目录配额、任务清理按钮、可选删除源视频 |
+| OpenCV 编码器不可用 | 无法生成 H.264 MP4 | 检测可用编码器，降级 `mp4v` 并记录到元数据 |
+
+## 16. 后续可扩展方向
+
+1. 增加 WebSocket 流式推理，减少 HTTP multipart 开销。
+2. 增加视频片段离线分析任务，输出时间轴事件。
+3. 增加目标跟踪，在检测低帧率下保持框连续。
+4. 增加 ROI 选择，只对画面局部做高分辨率推理。
+5. 增加模型 profile 切换：铁路模型、通用模型、CPU 快速验证、GPU 高精度。
+6. 增加当前帧截图和标注结果导出，供样本回流或模型测试工作台复用。
+7. 增加批量视频导出队列，对多个本地视频连续生成带检测/分割结果的本地文件。
+
+## 17. 推荐首版落地结论
+
+首版采用“浏览器本地播放视频 + Canvas 抽帧 + FastAPI 单帧推理 + ONNX Runtime GPU + Canvas 结果叠加”的实时预览方案，并增加“完整结果视频生成”后台任务：用户确认后上传完整视频，后端逐帧或按步长推理，把检测框和分割掩膜渲染进 `annotated.mp4`，保存到 `visualization-demo/outputs/video-runs/<run_id>/`。
+
+模型优先复用项目已有的 `vision-detector` 与 `vision-segmenter` 注册位：检测使用 PP-YOLOE-SOD-s，分割使用 PP-LiteSeg-STDC1。考虑到本机已确认 RTX 3090，默认目标运行在 `CUDAExecutionProvider`；当模型制品或 GPU Provider 不可用时，页面必须显示降级原因。
+
+该方案工程改动小、与现有仓库贴合，并且能自然扩展到后续 WebSocket、目标跟踪、离线分析和样本回流。