模型、区域点、颜色从配置中读取

biz/prison/indoor_biz.py

@@ -5,9 +5,10 @@ import requests
 from biz.base_frame_processor import BaseFrameProcessorWorker
 from algorithm.common.npu_yolo_onnx_person_car_phone import YOLOv8_ONNX
 from yolox.tracker.byte_tracker import BYTETracker
+from common.constants import MODEL_ROOT_PATH
 
 # ========================= 走廊场景专属配置 =========================
-MODEL_PATH = 'YOLO_Weight/kanshousuo.onnx'  # 犯人检测onnx模型路径
+DETECT_MODEL_PATH = 'YOLO_Weight/kanshousuo.onnx'  # 犯人检测onnx模型路径
 INPUT_SIZE = 640  # 模型输入尺寸
 RTSP_FPS = 10  # 视频流目标FPS
 ALERT_PUSH_INTERVAL = 5  # 相同报警5秒内仅推送1次
@@ -15,24 +16,35 @@ ALERT_PUSH_URL = "http://123.57.151.210:10000/picenter/websocket/test/process"
 # 消失判定：中心点在ROI内消失后，持续无检测的帧数（1.0秒，可微调）
 ROI_LOST_FRAMES_THRESH = int(0.5 * RTSP_FPS)
 
-# ========================= 5个ROI区域配置（相对坐标，适配任意分辨率） =========================
-# 格式：{ROI名称: [[x1,y1], [x2,y2], ...], ...} （多边形顶点，顺/逆时针均可）
-# 相对坐标：x/y 0~1（0=左/上，1=右/下），可直接根据场景调整
-ROI_CONFIG = {
-    "left_door_1": [[0.195, 0.242], [0.265, 0.17], [0.3, 0.63] ,[0.248, 0.8]],  # 左侧1门ROI
-    "left_door_2": [[0.3, 0.1], [0.34, 0.08], [0.35, 0.43], [0.322, 0.52]],      # 左侧2门 ROI
-    "left_door_3": [[0.355, 0.06], [0.42, 0], [0.42, 0.18], [0.362, 0.36]],  # 左侧3门ROI
-    "right_door_1": [[0.735, 0.142], [0.81, 0.22], [0.78, 0.8], [0.715, 0.65]],      # 右侧1门 ROI
-    "right_door_2": [[0.65, 0.06], [0.7, 0.09], [0.69, 0.5], [0.65, 0.4]]    # 右侧2门ROI
+# ========================= 默认ROI区域配置（当config.yaml未配置时使用） =========================
+DEFAULT_DOOR_ROIS = {
+    "left_door_1": {
+        "points": [[0.195, 0.242], [0.265, 0.17], [0.3, 0.63], [0.248, 0.8]],
+        "color": [255, 0, 0]
+    }
 }
 
 # ==================================================================================
 class PrisonerDoorDetector:
     def __init__(self, params=None):
         self.params = params or {}
-        # 1. 加载YOLO模型（仅提取犯人检测结果）
+
+        # 0. 从params解析ROI配置，无则使用默认值
+        door_rois_config = self.params.get('door_rois', DEFAULT_DOOR_ROIS)
+        self.roi_config = {}
+        self.roi_colors = {}
+        for door_name, door_cfg in door_rois_config.items():
+            self.roi_config[door_name] = door_cfg['points']
+            self.roi_colors[door_name] = tuple(door_cfg['color'])
+
+        model_path = self.params.get('model_path')
+        if model_path:
+            full_model_path = f"{MODEL_ROOT_PATH}/{model_path}"
+        else:
+            full_model_path = DETECT_MODEL_PATH
+
         self.detector = YOLOv8_ONNX(
-            MODEL_PATH,
+            full_model_path,
             conf_threshold=0.5,  # 置信度阈值，可根据模型精度调整
             iou_threshold=0.45,  # IOU阈值
             input_size=INPUT_SIZE
@@ -70,9 +82,9 @@ class PrisonerDoorDetector:
 
     def _get_roi_abs(self, roi_name):
         """相对坐标转绝对像素坐标（适配当前帧分辨率，OpenCV要求int32）"""
-        if roi_name not in ROI_CONFIG:
+        if roi_name not in self.roi_config:
             return None
-        roi_rel = np.array(ROI_CONFIG[roi_name], dtype=np.float64)
+        roi_rel = np.array(self.roi_config[roi_name], dtype=np.float64)
         roi_abs = roi_rel * np.array([self.frame_width, self.frame_height])
         return roi_abs.astype(np.int32)
 
@@ -119,23 +131,19 @@ class PrisonerDoorDetector:
         self.frame_height, self.frame_width = frame.shape[:2]
         current_frame_alerts = []  # 本帧报警信息
 
-        # ========================= 1. 初始化ROI绝对坐标并绘制5个ROI =========================
-        roi_colors = {  # 各ROI绘制颜色（自定义区分）
-            "left_door_1": (255, 0, 0), "left_door_2": (0, 255, 0),
-            "left_door_3": (0, 0, 255), "right_door_1": (255, 255, 0),
-            "right_door_2": (255, 165, 0)
-        }
+        # ========================= 1. 初始化ROI绝对坐标并绘制ROI =========================
         self.roi_abs_cache.clear()
-        for roi_name, _ in ROI_CONFIG.items():
+        for roi_name in self.roi_config:
             roi_abs = self._get_roi_abs(roi_name)
             if roi_abs is None:
                 continue
             self.roi_abs_cache[roi_name] = roi_abs
             # 绘制ROI多边形（闭合）+ ROI名称标签
             roi_draw = roi_abs.reshape((-1, 1, 2))  # OpenCV绘制要求形状 (n,1,2)
-            cv2.polylines(frame, [roi_draw], isClosed=True, color=roi_colors[roi_name], thickness=2)
+            color = self.roi_colors.get(roi_name, (255, 255, 255))
+            cv2.polylines(frame, [roi_draw], isClosed=True, color=color, thickness=2)
             cv2.putText(frame, roi_name, (roi_abs[0][0], roi_abs[0][1] - 5),
-                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, roi_colors[roi_name], 2)
+                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
 
         # ========================= 2. 模型推理：仅提取犯人检测框 =========================
         detect_results = self.detector(frame)