diff --git a/npu_yolo_pose_onnx.py b/npu_yolo_pose_onnx.py new file mode 100644 index 0000000..899a5dc --- /dev/null +++ b/npu_yolo_pose_onnx.py @@ -0,0 +1,275 @@ +# npu_yolo_pose_onnx.py +# 修复要点： +# 1. 正确处理 YOLOv8 Pose anchor 输出（避免 40+ 人） +# 2. 关键点坐标正确逆 letterbox（减 padding 再除 ratio） +# 3. visibility 使用 sigmoid +# 4. NMS 后限制最大人数，保证工程稳定性 + +import cv2 +import numpy as np +import onnxruntime as ort + + + + +# ------------------------------------------------- +# Letterbox +# ------------------------------------------------- +def letterbox(img, new_shape=(1280, 1280), color=(114, 114, 114)): + shape = img.shape[:2] # h, w + r = min(new_shape[0] / shape[0], new_shape[1] / shape[1]) + + new_unpad = (int(round(shape[1] * r)), int(round(shape[0] * r))) + dw = new_shape[1] - new_unpad[0] + dh = new_shape[0] - new_unpad[1] + dw /= 2 + dh /= 2 + + if shape[::-1] != new_unpad: + img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR) + + top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1)) + left, right = int(round(dw - 0.1)), int(round(dw + 0.1)) + img = cv2.copyMakeBorder( + img, top, bottom, left, right, + cv2.BORDER_CONSTANT, value=color + ) + return img, r, (dw, dh) + + +# ------------------------------------------------- +# Pose Skeleton Definition (COCO-17) +# ------------------------------------------------- +POSE_SKELETON = [ + (16,14),(14,12),(17,15),(15,13),(12,13), + (6,12),(7,13),(6,7), + (6,8),(7,9),(8,10),(9,11), + (2,3),(1,2),(1,3), + (2,4),(3,5),(4,6),(5,7) +] + +POSE_SKELETON = [(a-1, b-1) for (a, b) in POSE_SKELETON] + +POSE_COLORS = [ + (255,0,0),(255,85,0),(255,170,0),(255,255,0), + (170,255,0),(85,255,0),(0,255,0), + (0,255,85),(0,255,170),(0,255,255), + (0,170,255),(0,85,255),(0,0,255), + (85,0,255),(170,0,255),(255,0,255),(255,0,170) +] + + +# ------------------------------------------------- +# YOLOv8 Pose ONNX +# ------------------------------------------------- +class YOLOv8_Pose_ONNX: + def __init__( + self, + onnx_path, + conf_threshold=0.6, # ★ 提高阈值，避免 anchor 噪声 + iou_threshold=0.45, + input_size=1280, + max_persons=5 # ★ 限制最大人数 + ): + providers = [ + ("CANNExecutionProvider", { + "device_id": 0, + "arena_extend_strategy": "kNextPowerOfTwo", + "npu_mem_limit": 16 * 1024 * 1024 * 1024, + "precision_mode": "allow_fp32_to_fp16", + "op_select_impl_mode": "high_precision", + "enable_cann_graph": True, + }), + "CUDAExecutionProvider", + "CPUExecutionProvider", + ] + + self.session = ort.InferenceSession(onnx_path, providers=providers) + + # 获取真实工作 provider + actual_providers = self.session.get_providers() + + print("YOLO Providers:", actual_providers) + + if "CANNExecutionProvider" in actual_providers: + print("[INFO] YOLO 使用 CANNExecutionProvider（昇腾）") + elif 'CUDAExecutionProvider' in actual_providers: + print("[INFO] YOLO 使用 CUDAExecutionProvider（NVIDIA GPU）") + else: + print("[INFO] YOLO 使用 CPUExecutionProvider（非昇腾环境）") + + self.conf_threshold = conf_threshold + self.iou_threshold = iou_threshold + self.max_persons = max_persons + + self.input_name = self.session.get_inputs()[0].name + self.input_size = (input_size, input_size) + print(f"模型输入名称: {self.input_name}") + print(f"模型输入形状: {self.session.get_inputs()[0].shape}") + print(f"模型输出形状: {self.session.get_outputs()[0].shape}") + + + def nms(self, boxes, scores, iou_threshold=0.45): + """ + boxes: [N,4] xyxy + scores: [N] + """ + x1 = boxes[:, 0] + y1 = boxes[:, 1] + x2 = boxes[:, 2] + y2 = boxes[:, 3] + + areas = (x2 - x1) * (y2 - y1) + order = scores.argsort()[::-1] + + keep = [] + while order.size > 0: + i = order[0] + keep.append(i) + xx1 = np.maximum(x1[i], x1[order[1:]]) + yy1 = np.maximum(y1[i], y1[order[1:]]) + xx2 = np.minimum(x2[i], x2[order[1:]]) + yy2 = np.minimum(y2[i], y2[order[1:]]) + + w = np.maximum(0.0, xx2 - xx1) + h = np.maximum(0.0, yy2 - yy1) + inter = w * h + ovr = inter / (areas[i] + areas[order[1:]] - inter) + + inds = np.where(ovr <= iou_threshold)[0] + order = order[inds + 1] + + # 限制最大人数 + if len(keep) >= self.max_persons: + break + + return np.array(keep, dtype=int) +# ------------------------------------------------- + def preprocess(self, img): + self.orig_shape = img.shape[:2] + img, self.ratio, (self.dw, self.dh) = letterbox(img, self.input_size) + img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) + img = img.transpose(2, 0, 1).astype(np.float32) / 255.0 + img = np.expand_dims(img, axis=0) + return img + + def postprocess(self, preds, im0_shape): + """ + preds: onnx output, shape = [1, 56, 33600] + im0_shape: (h, w) of original frame + """ + + preds = preds[0] # [56, 33600] + preds = preds.transpose(1, 0) # [33600, 56] + + # ============================= + # 1. 拆分输出 + # ============================= + boxes = preds[:, 0:4] # cx, cy, w, h (input scale) + scores = preds[:, 4] # obj conf + kpts_raw = preds[:, 5:] # [33600, 51] = 17*3 + + # ============================= + # 2. 置信度筛选 + # ============================= + mask = scores > self.conf_threshold + boxes = boxes[mask] + scores = scores[mask] + kpts_raw = kpts_raw[mask] + + if boxes.shape[0] == 0: + return [] + + # ============================= + # 3. bbox cxcywh -> xyxy（input scale） + # ============================= + boxes_xyxy = np.zeros_like(boxes) + boxes_xyxy[:, 0] = boxes[:, 0] - boxes[:, 2] / 2 # x1 + boxes_xyxy[:, 1] = boxes[:, 1] - boxes[:, 3] / 2 # y1 + boxes_xyxy[:, 2] = boxes[:, 0] + boxes[:, 2] / 2 # x2 + boxes_xyxy[:, 3] = boxes[:, 1] + boxes[:, 3] / 2 # y2 + + # ============================= + # 4. inverse letterbox（bbox） + # ============================= + boxes_xyxy[:, [0, 2]] = (boxes_xyxy[:, [0, 2]] - self.dw) / self.ratio + boxes_xyxy[:, [1, 3]] = (boxes_xyxy[:, [1, 3]] - self.dh) / self.ratio + + boxes_xyxy[:, 0] = np.clip(boxes_xyxy[:, 0], 0, im0_shape[1]) + boxes_xyxy[:, 1] = np.clip(boxes_xyxy[:, 1], 0, im0_shape[0]) + boxes_xyxy[:, 2] = np.clip(boxes_xyxy[:, 2], 0, im0_shape[1]) + boxes_xyxy[:, 3] = np.clip(boxes_xyxy[:, 3], 0, im0_shape[0]) + + # ============================= + # 5. NMS + # ============================= + keep = self.nms(boxes_xyxy, scores, self.iou_threshold) + boxes_xyxy = boxes_xyxy[keep] + scores = scores[keep] + kpts_raw = kpts_raw[keep] + + # ============================= + # 6. 逐人处理 keypoints（关键） + # ============================= + results = [] + + for i in range(len(boxes_xyxy)): + x1, y1, x2, y2 = boxes_xyxy[i] + + # (51,) -> (17,3) + kpts = kpts_raw[i].reshape(17, 3).copy() + + + kpts[:, 0] = (kpts[:, 0] - self.dw) / self.ratio + kpts[:, 1] = (kpts[:, 1] - self.dh) / self.ratio + + # ✅ 加回 bbox offset（核心修复点） + #kpts[:, 0] += x1 + #kpts[:, 1] += y1 + + # clip + kpts[:, 0] = np.clip(kpts[:, 0], 0, im0_shape[1]) + kpts[:, 1] = np.clip(kpts[:, 1], 0, im0_shape[0]) + + # visibility sigmoid（防溢出） + kpts[:, 2] = 1.0 / (1.0 + np.exp(-np.clip(kpts[:, 2], -50, 50))) + + results.append({ + "bbox": [float(x1), float(y1), float(x2), float(y2)], + "conf": float(scores[i]), + "kpts": kpts + }) + + return results + + # ------------------------------------------------- + def __call__(self, frame): + inp = self.preprocess(frame) + pred = self.session.run(None, {self.input_name: inp})[0] + return self.postprocess(pred, frame.shape[:2]) + + @staticmethod + def draw_keypoints(frame, pose_results, vis_thres=0.3): + for res in pose_results: + kpts = res.get("kpts", None) # 注意这里对应 postprocess 返回的 key + if kpts is None or len(kpts) != 17: + continue + # 如果是 ndarray，转换为 list + if isinstance(kpts, np.ndarray): + kpts = kpts.tolist() + + for i, (x, y, v) in enumerate(kpts): + if v > vis_thres: + cv2.circle(frame, (int(x), int(y)), 5, POSE_COLORS[i], -1) + + for a, b in POSE_SKELETON: + if kpts[a][2] > vis_thres and kpts[b][2] > vis_thres: + cv2.line( + frame, + (int(kpts[a][0]), int(kpts[a][1])), + (int(kpts[b][0]), int(kpts[b][1])), + POSE_COLORS[a], + 2 + ) + return frame + diff --git a/rtsp_service_ws_kadian.py b/rtsp_service_ws_kadian.py new file mode 100644 index 0000000..e1b29ae --- /dev/null +++ b/rtsp_service_ws_kadian.py @@ -0,0 +1,1072 @@ +# rtsp_service_kadian.py +# 融合 Kadian_Detect_1221.py + rtsp_service_ws.py +# 支持多路RTSP、抽帧、分段保存MP4、WebSocket推送图像与告警 + +import cv2 +import numpy as np +import os +import time +import threading +import queue +import yaml +import json +import base64 +import asyncio +import requests +import websockets +from dataclasses import dataclass +from typing import Dict, Any, Tuple, List +from datetime import datetime +from test_cam import get_camera_preview_url + +# -------------------------- Kadian 检测相关导入 -------------------------- +from npu_yolo_onnx_person_car_phone import YOLOv8_ONNX # 主检测模型（人/车/后备箱/手机） +from npu_yolo_pose_onnx import YOLOv8_Pose_ONNX # Pose 专用模型 +from yolox.tracker.byte_tracker import BYTETracker + + +# ========================= 配置区 ========================= +# Kadian 模型路径与ROI（可根据实际情况修改） +DETECT_MODEL_PATH = 'YOLO_Weight/car_opentrunk_person_phone.onnx' +POSE_MODEL_PATH = 'YOLO_Weight/yolov8l-pose.onnx' + +# 默认相对ROI（与原文件一致） +#ROI_RELATIVE = np.array([ +# [0.10989583333333333, 0.006481481481481481], +# [0.421875, 0.005555555555555556], +# [0.9921875, 0.9888888888888889], +# [0.3411458333333333, 0.9861111111111112] +#]) + + +ROI_RELATIVE=np.array([ + [0.15,0.001], + [0.5,0.001], + [1.0,0.8], + [0.35,1.0] +]) + +ALERT_PUSH_INTERVAL = 5.0 + +# 输入尺寸 +PERSON_CAR_INPUT_SIZE = 640 +POSE_INPUT_SIZE = 640 + +# RTSP 服务配置 +RTSP_TARGET_FPS = 10.0 +WS_HOST = "0.0.0.0" +WS_PORT = 8765 +WS_PORT_2 = 8764 # 新增：第二个WebSocket端口 + +# WebSocket 客户端集合 +ws_clients = set() +ws_clients_2 = set() # 新增：第二个WebSocket客户端集合 + + +# ========================= 数据结构 ========================= +@dataclass +class CameraConfig: + id: int + name: str + index: str + rtsp_url: str + + +# ========================= Kadian TrafficMonitor（精简版，专为服务设计） ========================= +class KadianDetector: + def __init__(self, roi_points=ROI_RELATIVE): + # 模型加载 + self.detector = YOLOv8_ONNX(DETECT_MODEL_PATH, conf_threshold=0.25, iou_threshold=0.45, + input_size=PERSON_CAR_INPUT_SIZE) + self.pose_detector = YOLOv8_Pose_ONNX(POSE_MODEL_PATH, conf_threshold=0.7, iou_threshold=0.6, + input_size=POSE_INPUT_SIZE) + + # Tracker + class TrackerArgs: + track_thresh = 0.25 + track_buffer = 30 + match_thresh = 0.8 + mot20 = False + self.tracker = BYTETracker(TrackerArgs(), frame_rate=10.0) + + self.track_role = {} + + self.fps = RTSP_TARGET_FPS + + # ROI 处理（支持相对/绝对） + #self.roi_points = roi_points.astype(np.int32) + self.roi_points = np.array(roi_points, dtype=np.float64) if roi_points is not None else None + + # ========================================== + # 超参数设置 (Hyperparameters) + # ========================================== + + # 1. 业务判定时间阈值 + self.TIME_THRESHOLD_ONLY_ONE = 3.0 # 单人单检判定时长 + self.TIME_THRESHOLD_NOBODY = 2.0 # 无人检查判定时长 + + # 后备箱检查判定阈值 + self.TIME_THRESHOLD_TRUNK_OPEN = 0.5 + + # 新增：手机检测判定阈值 + self.TIME_THRESHOLD_PHONE = 1.0 # 手机检测持续1秒（30帧 @30fps） + self.TIME_TOLERANCE_PHONE = 0.5 # 手机丢失缓冲时间（防抖动） + + # 新增：制服检测判定阈值 + self.TIME_THRESHOLD_UNIFORM = 1.0 # 制服不合规判定时长 + self.TIME_TOLERANCE_UNIFORM = 0.5 # 制服合规恢复缓冲时间 + + # 车辆最小停留时间阈值 (小于此时间视为无人检查/直接通过) + self.TIME_THRESHOLD_CAR_MIN_DURATION = 3.0 + + # 2. Person 丢帧缓冲 + self.TIME_TOLERANCE_PERSON = 1.0 + + # 3. Car 丢帧/ID维持缓冲 + self.TIME_TOLERANCE_CAR = 0.5 + + # --- 计算对应的帧数阈值 --- + self.frame_thresh_one = int(self.TIME_THRESHOLD_ONLY_ONE * self.fps) + self.frame_thresh_nobody = int(self.TIME_THRESHOLD_NOBODY * self.fps) + self.frame_thresh_trunk_valid = int(self.TIME_THRESHOLD_TRUNK_OPEN * self.fps) + + # 新增：手机检测帧数阈值 + self.frame_thresh_phone = int(self.TIME_THRESHOLD_PHONE * self.fps) + self.frame_buffer_phone = int(self.TIME_TOLERANCE_PHONE * self.fps) + + # 新增：制服检测帧数阈值 + self.frame_thresh_uniform = int(self.TIME_THRESHOLD_UNIFORM * self.fps) + self.frame_buffer_uniform = int(self.TIME_TOLERANCE_UNIFORM * self.fps) + + self.frame_thresh_car_min_duration = int(self.TIME_THRESHOLD_CAR_MIN_DURATION * self.fps) + + self.frame_buffer_limit_person = int(self.TIME_TOLERANCE_PERSON * self.fps) + self.frame_buffer_limit_car = int(self.TIME_TOLERANCE_CAR * self.fps) + + print(f"\n超参数设置:") + print(f" FPS: {self.fps:.2f}") + print(f" 判定 'Only One' / 'Nobody' 需连续: {self.frame_thresh_one} 帧") + print(f" 判定 'Trunk Checked' 需累计检测: {self.frame_thresh_trunk_valid} 帧") + print(f" 判定 'Phone Detected' 需累计检测: {self.frame_thresh_phone} 帧") + print(f" 手机丢失缓冲帧数: {self.frame_buffer_phone} 帧") + print(f" 判定 'Uniform Invalid' 需连续检测: {self.frame_thresh_uniform} 帧") + print(f" 制服合规恢复缓冲帧数: {self.frame_buffer_uniform} 帧") + print(f" 判定 'Too Fast' (视为Nobody) 最小停留: {self.frame_thresh_car_min_duration} 帧") + + + self.current_frame_idx = 0 + self.cnt_frame_one_person = 0 + self.cnt_frame_nobody = 0 + self.cnt_missing_buffer_person = 0 + + # 手机检测状态变量（独立于车辆） + self.phone_detection_frames = 0 # 连续检测到手机的帧数 + self.phone_missing_frames = 0 # 连续未检测到手机的帧数 + self.phone_alert_active = False # 手机报警是否激活 + + # 新增：制服检测状态变量 + self.pose_person_count = 0 # 骨骼点模型检测的ROI内人员数量 + self.uniform_alert_active = False # 制服报警是否激活 + self.uniform_detection_frames = 0 # 连续检测到制服不合规的帧数 + self.uniform_recovery_frames = 0 # 连续恢复合规的帧数 + + # 车辆注册表 (字典) + self.roi_car_registry = {} + + # 违规车辆记录 (后备箱未检) + self.unchecked_trunk_alerts = {} + + # 违规车辆记录 (通过过快 -> 归类为 Nobody) + self.fast_pass_alerts = {} + + def _get_roi_points(self, frame_width: int, frame_height: int): + """ + 每帧动态计算正确的 ROI 绝对坐标，并确保类型为 np.int32 + 用于 pointPolygonTest 和 polylines + """ + if self.roi_points is None: + # 使用默认相对坐标 + default_rel = np.array([ + [0.15, 0.01], + [0.45, 0.01], + [0.95, 0.95], + [0.35, 0.95] + ], dtype=np.float64) + roi_abs = default_rel * np.array([frame_width, frame_height]) + else: + if self.roi_points.max() <= 1.0: + # 相对坐标 → 转换为绝对 + roi_abs = self.roi_points * np.array([frame_width, frame_height]) + else: + # 绝对坐标，直接使用 + roi_abs = self.roi_points.copy() + + # 强制转为 int32（关键！解决 OpenCV 断言错误） + return roi_abs.astype(np.int32) + + def check_point_in_roi(self,roi_points, point): + return cv2.pointPolygonTest(roi_points, point, False) >= 0 + + def compute_iou(self,boxA, boxB): + # box = [x1, y1, x2, y2] + xA = max(boxA[0], boxB[0]) + yA = max(boxA[1], boxB[1]) + xB = min(boxA[2], boxB[2]) + yB = min(boxA[3], boxB[3]) + + interW = max(0, xB - xA) + interH = max(0, yB - yA) + interArea = interW * interH + + boxAArea = (boxA[2] - boxA[0]) * (boxA[3] - boxA[1]) + boxBArea = (boxB[2] - boxB[0]) * (boxB[3] - boxB[1]) + + unionArea = boxAArea + boxBArea - interArea + if unionArea == 0: + return 0.0 + + return interArea / unionArea + + def draw_alert(self, frame, text, color=(0, 0, 255), sub_text=None, offset_y=0): + """在右上角绘制警告文字 (支持垂直偏移，防止文字重叠)""" + font_scale = 1.5 + thickness = 3 + font = cv2.FONT_HERSHEY_SIMPLEX + + (text_w, text_h), _ = cv2.getTextSize(text, font, font_scale, thickness) + x = self.width - text_w - 20 + y = 50 + text_h + offset_y # 增加 Y 轴偏移 + + cv2.rectangle(frame, (x - 10, y - text_h - 10), (x + text_w + 10, y + 10), (0, 0, 0), -1) + cv2.putText(frame, text, (x, y), font, font_scale, color, thickness) + + if sub_text: + cv2.putText(frame, sub_text, (x, y + 40), font, 0.7, (200, 200, 200), 2) + + def is_point_in_box(self, point, box): + px, py = point + x1, y1, x2, y2 = box + return x1 < px < x2 and y1 < py < y2 + + + def process_frame(self, frame, camera_id: int, timestamp: float) -> Dict[str, Any]: + h, w = frame.shape[:2] + self.width, self.height = w, h + + self.current_frame_idx += 1 + # ========= 每帧动态获取正确的 ROI（int32）========= + roi_points_int32 = self._get_roi_points(w, h) # shape: (4, 2), dtype: int32 + roi_points_draw = roi_points_int32.reshape((-1, 1, 2)) # shape: (4, 1, 2) 用于绘制 + + + current_time_sec = timestamp + + pose_results = self.pose_detector(frame) + + # ========= 主检测 ========= + detections = self.detector(frame) + + dets_xyxy = [] + dets_roles = [] + dets_for_tracker = [] + + # ========= 当前帧所有警告列表（关键改动）========== + current_frame_alerts = [] # 每帧清空，重新收集 + + if detections: + for det in detections: + x1, y1, x2, y2, conf, cls_id = det # x1, y1, x2, y2为角点坐标，x1 y1为左上角，x2 y2为右下角 + dets_xyxy.append([x1, y1, x2, y2]) + dets_for_tracker.append([x1, y1, x2, y2, conf]) + if cls_id == 0: + dets_roles.append("car") + + elif cls_id == 1: + dets_roles.append("opentrunk") + + elif cls_id == 2: + dets_roles.append("person") + + elif cls_id == 3: + dets_roles.append("phone") + # print(f'dets_roles: {dets_roles}') + + dets = np.array(dets_for_tracker, dtype=np.float32) if len(dets_for_tracker) else np.empty((0, 5)) + + tracks = self.tracker.update( + dets, + [self.height, self.width], + [self.height, self.width] + ) + # print("tracks: {}".format(tracks)) + # 绘制骨骼 + frame = YOLOv8_Pose_ONNX.draw_keypoints(frame, pose_results) + # ========= 绘制 ROI ========= + cv2.polylines(frame, [roi_points_draw], isClosed=True, color=(255, 0, 0), thickness=3) + + # ========= 单帧统计变量 ========= + current_roi_person_count = 0 + current_roi_trunk_count = 0 + current_roi_phone_count = 0 + + # 临时存储本帧的目标，用于后续关联分析 + current_cars = [] # {'id':, 'box':} + current_trunks = [] # (cx, cy) + + for t in tracks: + # print("t: {}".format(t)) + tid = t.track_id + # cls_id = -1 + + # IoU 匹配角色 + # if tid not in track_role and dets_xyxy: + REVALIDATE_FRAME_INTERVAL = 10 + #if tid not in self.track_role: + if (self.current_frame_idx % REVALIDATE_FRAME_INTERVAL == 0) or (tid not in self.track_role): + best_iou = 0 + best_role = "unknown" + + t_box = list(map(float, t.tlbr)) # [x1,y1,x2,y2] + + for i, box in enumerate(dets_xyxy): + iou_val = self.compute_iou(t_box, box) + if iou_val > best_iou: + best_iou = iou_val + best_role = dets_roles[i] + if best_iou > 0.1: + self.track_role[tid] = best_role + else: + self.track_role[tid] = "unknown" + + role = self.track_role.get(tid, "unknown") + cls_id = -1 + if role == "car": + cls_id = 0 + elif role == "opentrunk": + cls_id = 1 + elif role == "person": + cls_id = 2 + elif role == "phone": + cls_id = 3 + # print("tid: {}, role: {}, cls: {}".format(tid, role,cls_id)) + + x1, y1, x2, y2 = map(int, t.tlbr) + + cx, cy = (x1 + x2) // 2, (y1 + y2) // 2 + + color = None + label = None + + if self.check_point_in_roi(roi_points_int32,(cx, cy)): + if cls_id == 0: # Car + color = (0, 255, 0) + + current_cars.append({'id': tid, 'box': [x1, y1, x2, y2]}) + + if tid not in self.roi_car_registry: + self.roi_car_registry[tid] = { + 'first_seen': self.current_frame_idx, + 'last_seen': self.current_frame_idx, + 'trunk_frames': 0, + 'is_checked': False, + } + else: + self.roi_car_registry[tid]['last_seen'] = self.current_frame_idx + + label = f"Car:{tid} IN" + + elif cls_id == 1: # Opentrunk + current_roi_trunk_count += 1 + color = (255, 165, 0) + current_trunks.append((cx, cy)) + label = "OpenTrunk IN" + + elif cls_id == 2: # Person + current_roi_person_count += 1 + color = (255, 0, 255) + label = "Person IN" + + elif cls_id == 3: # Phone（主模型已支持） + current_roi_phone_count += 1 + color = (0, 0, 139) + + else: + color = (255, 255, 255) + label = "Unknown" + + # label = f"ID:{tid} IN" + + # 特殊显示: 如果这辆车已经合格，框变蓝色 + if cls_id == 0 and tid in self.roi_car_registry and self.roi_car_registry[tid][ + 'is_checked']: + color = (255, 255, 0) # Cyan for checked cars + label += " (Checked)" + else: + color = (0, 0, 255) + label = "OUT" + + cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2) + cv2.putText(frame, label, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2) + + # ========================================== + # 4. 从骨骼点模型中统计ROI内人员数量 + # ========================================== + self.pose_person_count = 0 + # if pose_results[0].boxes is not None: + # pose_boxes = pose_results[0].boxes + # for box in pose_boxes: + # # 获取人体检测框的中心点 + # x1, y1, x2, y2 = map(int, box.xyxy[0]) + # cx, cy = (x1 + x2) // 2, (y1 + y2) // 2 + # + # # 判断中心点是否在ROI内 + # if self.check_point_in_roi((cx, cy)): + # self.pose_person_count += 1 + + if pose_results: + for pose in pose_results: + + x1, y1, x2, y2 = pose['bbox'][0], pose['bbox'][1], pose['bbox'][2], pose['bbox'][3] + cx, cy = (x1 + x2) // 2, (y1 + y2) // 2 + # 判断中心点是否在ROI内 + if self.check_point_in_roi(roi_points_int32,(cx, cy)): + self.pose_person_count += 1 + + # ========================================== + # 5. 关联分析: 哪个后备箱属于哪辆车? + # ========================================== + for car_info in current_cars: + c_id = car_info['id'] + c_box = car_info['box'] + + trunk_found_for_this_car = False + for t_pt in current_trunks: + if self.is_point_in_box(t_pt, c_box): + trunk_found_for_this_car = True + break + + if trunk_found_for_this_car: + self.roi_car_registry[c_id]['trunk_frames'] += 1 + if self.roi_car_registry[c_id]['trunk_frames'] >= self.frame_thresh_trunk_valid: + self.roi_car_registry[c_id]['is_checked'] = True + + # ========================================== + # 6. 独立的手机检测逻辑（不与车辆绑定） + # ========================================== + if current_roi_phone_count > 0: + # 检测到手机框 + self.phone_detection_frames += 1 + self.phone_missing_frames = 0 # 重置丢失计数器 + + # 当检测累计达到阈值时，激活报警 + if self.phone_detection_frames >= self.frame_thresh_phone: + self.phone_alert_active = True + else: + # 未检测到手机框 + self.phone_missing_frames += 1 + + # 如果之前检测到手机，重置检测计数器 + if self.phone_detection_frames > 0: + # 只有在连续丢失超过缓冲帧数时才重置 + if self.phone_missing_frames >= self.frame_buffer_phone: + self.phone_detection_frames = 0 + self.phone_alert_active = False + else: + # 从未检测到手机，保持状态 + pass + + # ========================================== + # 7. 制服检测逻辑（比较两个模型的人员数量） + # ========================================== + # 比较骨骼点模型和业务检测模型的人员数量 + uniform_invalid = False + + if self.pose_person_count > current_roi_person_count: + # 骨骼点模型检测到的人员多于业务检测模型 + # 说明有人没穿执勤制服 + uniform_invalid = True + self.uniform_detection_frames += 1 + self.uniform_recovery_frames = 0 # 重置恢复计数器 + + # 当连续检测不合规达到阈值时，激活报警 + if self.uniform_detection_frames >= self.frame_thresh_uniform: + self.uniform_alert_active = True + else: + # 人员数量匹配或业务模型检测更多（理论上不会） + self.uniform_recovery_frames += 1 + + # 如果之前有不合规检测，检查是否需要关闭报警 + if self.uniform_detection_frames > 0: + # 只有在连续合规超过缓冲帧数时才重置 + if self.uniform_recovery_frames >= self.frame_buffer_uniform: + self.uniform_detection_frames = 0 + self.uniform_alert_active = False + else: + # 从未检测到不合规，保持状态 + pass + + # ========================================== + # 8. 维护车辆注册表 & 生成离场报警 + # ========================================== + active_car_ids = [] + cars_to_remove = [] + + for car_id, info in self.roi_car_registry.items(): + last_seen = info['last_seen'] + + if (self.current_frame_idx - last_seen) <= self.frame_buffer_limit_car: + active_car_ids.append(car_id) + else: + cars_to_remove.append(car_id) + + # 执行删除 并 检查违规 + for car_id in cars_to_remove: + car_info = self.roi_car_registry[car_id] + + duration_frames = car_info['last_seen'] - car_info['first_seen'] + + # 情况1：通过时间太短 -> 归类为 Nobody (Too Fast) + if duration_frames < self.frame_thresh_car_min_duration: + print(f"ALARM: Car {car_id} passed too fast -> Regarded as Nobody Checked!") + self.fast_pass_alerts[car_id] = self.current_frame_idx + int(3.0 * self.fps) + + # 情况2：时间够长，但没检查后备箱 -> Unchecked Trunk + elif not car_info['is_checked']: + print(f"ALARM: Car {car_id} left without checking trunk!") + self.unchecked_trunk_alerts[car_id] = self.current_frame_idx + int(3.0 * self.fps) + + del self.roi_car_registry[car_id] + + effective_car_count = len(active_car_ids) + + # ========================================== + # 9. 业务逻辑判定 (Only One / Nobody) + # ========================================== + status_text = "" + + if effective_car_count > 0: + # --- Only One --- + if current_roi_person_count == 1: + self.cnt_frame_one_person += 1 + self.cnt_missing_buffer_person = 0 + self.cnt_frame_nobody = 0 + + # --- Nobody --- + elif current_roi_person_count == 0: + if self.cnt_frame_one_person > 0 and self.cnt_missing_buffer_person < self.frame_buffer_limit_person: + self.cnt_frame_one_person += 1 + self.cnt_missing_buffer_person += 1 + self.cnt_frame_nobody = 0 + status_text = f"Person Buffer ({self.cnt_missing_buffer_person}/{self.frame_buffer_limit_person})" + else: + self.cnt_frame_one_person = 0 + self.cnt_missing_buffer_person = 0 + self.cnt_frame_nobody += 1 + else: + self.cnt_frame_one_person = 0 + self.cnt_missing_buffer_person = 0 + self.cnt_frame_nobody = 0 + else: + self.cnt_frame_one_person = 0 + self.cnt_missing_buffer_person = 0 + self.cnt_frame_nobody = 0 + + # ========================================== + # 10. 显示报警 (UI分层优化) + # ========================================== + + # 更新调试信息，包含所有检测状态 + phone_status = f"Phone: {current_roi_phone_count}" + if self.phone_alert_active: + phone_status += " (ALERT)" + + uniform_status = f"Uniform: Pose={self.pose_person_count}, Model={current_roi_person_count}" + if self.uniform_alert_active: + uniform_status += " (INVALID!)" + + debug_info = f"Cars: {len(active_car_ids)} | Person: {current_roi_person_count} | Trunk: {current_roi_trunk_count} | {phone_status}" + cv2.putText(frame, debug_info, (20, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2) + cv2.putText(frame, uniform_status, (20, 70), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2) + + # 使用 offset 实现报警堆叠，防止遮挡 + alert_offset = 0 + + # 第一层：实时状态 (Real-time Status) + # ------------------------------------------------ + # A. 显示 Only One + if self.cnt_frame_one_person >= self.frame_thresh_one: + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Only One", + } + ) + self.draw_alert(frame, "Only One", (0, 255, 255), status_text, offset_y=alert_offset) + alert_offset += 100 + + # B. 显示 Nobody (实时状态) + elif self.cnt_frame_nobody >= self.frame_thresh_nobody: + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Nobody", + } + ) + self.draw_alert(frame, "Nobody", (0, 0, 255), offset_y=alert_offset) + alert_offset += 100 + + # C. 显示 Trunk Checked (在车辆存活期间) + for car_id in active_car_ids: + if car_id in self.roi_car_registry and self.roi_car_registry[car_id]['is_checked']: + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Trunk Checked", + } + ) + self.draw_alert(frame, "Trunk Checked!!", (0, 255, 0), offset_y=alert_offset) + alert_offset += 100 + break # 只显示一次 + + # D. 显示 Playing Phone（独立检测，不与车辆绑定） + if self.phone_alert_active: + # 可以显示检测的持续时间 + duration_seconds = self.phone_detection_frames / self.fps + sub_text = f"Detected for {duration_seconds:.1f}s" + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Playing Phone", + } + ) + self.draw_alert(frame, "Playing Phone", (255, 0, 0), sub_text, offset_y=alert_offset) + alert_offset += 100 + + # E. 新增：显示 Unvaild Uniform!! + if self.uniform_alert_active: + # 显示具体数量差异 + diff = self.pose_person_count - current_roi_person_count + sub_text = f"Missing {diff} uniform(s)" + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Unvaild Uniform!!", + } + ) + self.draw_alert(frame, "Unvaild Uniform!!", (255, 165, 0), sub_text, offset_y=alert_offset) + alert_offset += 100 + + # 第二层：离场违规 (Post-Event Alerts) + # ------------------------------------------------ + + # F. 显示 Unchecked Trunk + expired_alerts = [cid for cid, end_frame in self.unchecked_trunk_alerts.items() if + self.current_frame_idx > end_frame] + for cid in expired_alerts: + del self.unchecked_trunk_alerts[cid] + + if len(self.unchecked_trunk_alerts) > 0: + alert_text = f"Unchecked Trunk! (ID:{list(self.unchecked_trunk_alerts.keys())})" + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Unchecked Trunk", + } + ) + self.draw_alert(frame, alert_text, (0, 0, 255), offset_y=alert_offset) + alert_offset += 100 + + # G. 显示 Nobody (离场结果) + expired_fast_alerts = [cid for cid, end_frame in self.fast_pass_alerts.items() if + self.current_frame_idx > end_frame] + for cid in expired_fast_alerts: + del self.fast_pass_alerts[cid] + + if len(self.fast_pass_alerts) > 0: + alert_text = f"Nobody (ID:{list(self.fast_pass_alerts.keys())})" + current_frame_alerts.append( + { + 'time': current_time_sec, + 'action': "Nobody", + } + ) + self.draw_alert(frame, alert_text, (0, 0, 255), offset_y=alert_offset) + alert_offset += 100 + + return { + "image": frame, + + "alerts":current_frame_alerts + } + +# ========================= WebSocket 服务线程 ========================= +class WebSocketSender(threading.Thread): + def __init__(self, send_queue: queue.Queue, stop_event: threading.Event): + super().__init__(daemon=True) + self.send_queue = send_queue + self.stop_event = stop_event + + async def _ws_handler(self, websocket): + ws_clients.add(websocket) + try: + async for _ in websocket: + pass + finally: + ws_clients.discard(websocket) + + async def _broadcaster(self): + while not self.stop_event.is_set(): + try: + msg = await asyncio.to_thread(self.send_queue.get, timeout=0.5) + except queue.Empty: + continue + data = json.dumps(msg) + dead = [] + for ws in list(ws_clients): + try: + await ws.send(data) + except: + dead.append(ws) + for ws in dead: + ws_clients.discard(ws) + self.send_queue.task_done() + + async def _run_async(self): + async with websockets.serve(self._ws_handler, WS_HOST, WS_PORT): + print(f"[INFO] WebSocket server started at ws://{WS_HOST}:{WS_PORT}") + await self._broadcaster() + + def run(self): + asyncio.run(self._run_async()) + + +# ========================= WebSocket 服务线程2 ========================= +class WebSocketSender2(threading.Thread): + def __init__(self, send_queue: queue.Queue, stop_event: threading.Event): + super().__init__(daemon=True) + self.send_queue = send_queue + self.stop_event = stop_event + + async def _ws_handler(self, websocket): + ws_clients_2.add(websocket) + try: + async for _ in websocket: + pass + finally: + ws_clients_2.discard(websocket) + + async def _broadcaster(self): + while not self.stop_event.is_set(): + try: + msg = await asyncio.to_thread(self.send_queue.get, timeout=0.5) + except queue.Empty: + continue + data = json.dumps(msg) + dead = [] + for ws in list(ws_clients_2): + try: + await ws.send(data) + except: + dead.append(ws) + for ws in dead: + ws_clients_2.discard(ws) + self.send_queue.task_done() + + async def _run_async(self): + async with websockets.serve(self._ws_handler, WS_HOST, WS_PORT_2): + print(f"[INFO] WebSocket server 2 started at ws://{WS_HOST}:{WS_PORT_2}") + await self._broadcaster() + + def run(self): + asyncio.run(self._run_async()) + + +# ========================= RTSP 抓流线程 ========================= +class RTSPCaptureWorker(threading.Thread): + def __init__(self, camera_cfg: CameraConfig, raw_queue: queue.Queue, stop_event: threading.Event): + super().__init__(daemon=True) + self.camera_cfg = camera_cfg + self.raw_queue = raw_queue + self.stop_event = stop_event + # 添加重连计数器 + self.reconnect_count = 0 + self.max_reconnects = 5 + self.rtsp_url = "" + + def run(self): + while not self.stop_event.is_set(): + try: + + if self.reconnect_count >= self.max_reconnects: + print(f"[WARN] RTSP: {self.camera_cfg.name} reach max reconnects, refresh url") + self.reconnect_count = 0 + new_url = self.refresh_video_url() + if new_url: + self.rtsp_url = new_url + else: + print(f"[ERROR] refresh RTSP URL is empty, do nothing") + + # 检查rtsp_url是否为空或None，如果是则重新获取 + if not self.rtsp_url: + print(f"[WARN] RTSP URL is empty, refreshing...") + new_url = self.refresh_video_url() + if new_url: + self.rtsp_url = new_url + else: + print(f"[ERROR] RTSP URL is still empty, retrying in 5 seconds") + time.sleep(5) + continue + + # 方法1：使用TCP传输（更稳定） + rtsp_url = self.rtsp_url + if "?" not in rtsp_url: + rtsp_url += "?transport=tcp" # 强制TCP传输 + else: + rtsp_url += "&transport=tcp" + + # 方法2：添加更多FFmpeg参数 + cap = cv2.VideoCapture(rtsp_url, cv2.CAP_FFMPEG) + + # 方法3：设置缓冲区大小 + cap.set(cv2.CAP_PROP_BUFFERSIZE, 10) # 增加缓冲区 + + # 方法4：设置超时和重连参数 + os.environ["OPENCV_FFMPEG_CAPTURE_OPTIONS"] = \ + "rtsp_transport;tcp|buffer_size;1024000|max_delay;500000|stimeout;2000000" + + # 方法5：设置解码器flags，忽略解码错误 + # cap.set(cv2.CAP_PROP_HW_ACCELERATION, cv2.VIDEO_ACCELERATION_ANY) + + if not cap.isOpened(): + print(f"[ERROR] Cannot open RTSP: {self.rtsp_url}") + time.sleep(2) + self.reconnect_count += 1 + continue + + print(f"[INFO] Successfully opened RTSP: {self.name}") + self.reconnect_count = 0 # 重置重连计数 + + # # 设置帧率（可选） + # cap.set(cv2.CAP_PROP_FPS, 25) + + while not self.stop_event.is_set(): + ret, frame = cap.read() + if not ret: + # 检查流是否结束 + print(f"[WARN] Failed to read frame from {self.camera_cfg.name}") + + # 检查是否还有数据 + time.sleep(0.1) + # 尝试几次后重连 + break + + item = { + "camera_id": self.camera_cfg.id, + "camera_name": self.camera_cfg.name, + "timestamp": time.time(), + "frame": frame, + } + + try: + # 添加队列满时的处理 + if self.raw_queue.full(): + # 丢弃最旧的一帧 + try: + self.raw_queue.get_nowait() + self.raw_queue.task_done() + except queue.Empty: + pass + + self.raw_queue.put(item, timeout=0.5) + except queue.Full: + print(f"[WARN] Queue full, dropping frame from {self.camera_cfg.name}") + continue + + # 控制读取速度，避免过快 + time.sleep(0.02) # 约50ms间隔 + + cap.release() + + except Exception as e: + print(f"[ERROR] Error in RTSP capture for {self.camera_cfg.name}: {e}") + time.sleep(2) + self.reconnect_count += 1 + + if self.reconnect_count >= self.max_reconnects: + print(f"[ERROR] Max reconnects reached for {self.camera_cfg.name}, stopping.") + + def refresh_video_url(self): + """ + 重新通过视频ID获取视频URL，调用test_cam.py中的get_camera_preview_url方法 + + 返回: + str: 新的视频URL，如果获取失败则返回None + """ + try: + + # 获取视频ID（camera_cfg.index） + video_id = self.camera_cfg.index + + # 调用test_cam.py中的函数 + result = get_camera_preview_url(video_id) + + # 解析结果（与test_cam.py相同） + if 'data' in result and 'url' in result['data']: + new_url = result['data']['url'] + print(f"[INFO] get rtsp url success, URL: {new_url}") + return new_url + else: + print(f"[ERROR] get rtsp url failed: {result}") + return None + + except Exception as e: + print(f"[ERROR] get rtsp url error: {str(e)}") + return None + + +# ========================= 帧处理线程 ========================= +class FrameProcessorWorker(threading.Thread): + def __init__(self, raw_queue: queue.Queue, ws_queue: queue.Queue, ws_queue_2: queue.Queue, stop_event: threading.Event): + super().__init__(daemon=True) + self.raw_queue = raw_queue + self.ws_queue = ws_queue + self.ws_queue_2 = ws_queue_2 # 新增：第二个WebSocket队列 + self.stop_event = stop_event + + self.last_ts: Dict[int, float] = {} + + # 每个摄像头一个独立的 Kadian 检测器实例 + self.kadian_detectors: Dict[int, KadianDetector] = {} + + self.last_alert_push_time: Dict[int,Dict[str,float]]={} + + + def _encode_base64(self, img): + _, buf = cv2.imencode(".jpg", img) + return base64.b64encode(buf).decode("ascii") + + def run(self): + target_interval = 1.0 / RTSP_TARGET_FPS + while not self.stop_event.is_set(): + try: + item = self.raw_queue.get(timeout=0.5) + except queue.Empty: + continue + + cam_id = item["camera_id"] + ts = item["timestamp"] + frame = item["frame"] + + # 抽帧控制 + if ts - self.last_ts.get(cam_id, 0) < target_interval: + self.raw_queue.task_done() + continue + self.last_ts[cam_id] = ts + + # 获取检测器实例 + if cam_id not in self.kadian_detectors: + self.kadian_detectors[cam_id] = KadianDetector() + detector = self.kadian_detectors[cam_id] + + # 执行检测 + result = detector.process_frame(frame.copy(), cam_id, ts) + + result_img = result["image"] + result_type = result["alerts"] + #print(f"alerts: {result_type}") + + # ========= 核心修改：过滤5秒内重复的action ========= + # 初始化当前摄像头的推送时间记录 + if cam_id not in self.last_alert_push_time: + self.last_alert_push_time[cam_id] = {} + + # 筛选出符合推送条件的action（5秒内未推送过） + push_actions = [] + current_time = time.time() + for alert in result_type: + action = alert['action'] + last_push = self.last_alert_push_time[cam_id].get(action, 0) + # 检查是否超过推送间隔 + if current_time - last_push >= ALERT_PUSH_INTERVAL: + push_actions.append(action) + # 更新该action的最后推送时间 + self.last_alert_push_time[cam_id][action] = current_time + + # 通过 WebSocket 发送帧结果 + try: + img_b64 = self._encode_base64(result_img) + except Exception as e: + print(f"[ERROR] Encode image failed: {e}") + img_b64 = None + + if img_b64 is not None: + # 将abnormal_actions对象数组转换为字符串数组 + #action_names = [action_info['action'] for action_info in push_actions] + + msg = { + "msg_type": "frame", + "camera_id": 0, + "timestamp": ts, + #"result_type": action_names, + "result_type": push_actions, + "image_base64": img_b64, + } + try: + self.ws_queue.put(msg, timeout=1.0) + if push_actions and len(push_actions) > 0: + self.ws_queue_2.put(msg, timeout=1.0) + except queue.Full: + print("[WARN] ws_send_queue full, drop frame message") + + self.raw_queue.task_done() + + +# ========================= 服务主类 ========================= +class RTSPService: + def __init__(self, config_path: str = "config.yaml"): + with open(config_path, "r", encoding="utf-8") as f: + cfg = yaml.safe_load(f) + self.cameras = [CameraConfig(id=c["id"], name=c.get("name", f"cam_{c['id']}"), index = c["index"], rtsp_url=c["rtsp_url"]) + for c in cfg.get("cameras", [])] + + self.stop_event = threading.Event() + self.raw_queue = queue.Queue(maxsize=500) + self.ws_queue = queue.Queue(maxsize=1000) + self.ws_queue_2 = queue.Queue(maxsize=1000) # 新增：第二个WebSocket队列 + + self.capture_workers = [] + self.processor = FrameProcessorWorker(self.raw_queue, self.ws_queue, self.ws_queue_2, self.stop_event) + self.ws_sender = WebSocketSender(self.ws_queue, self.stop_event) + self.ws_sender_2 = WebSocketSender2(self.ws_queue_2, self.stop_event) # 新增：第二个WebSocket发送器 + + def start(self): + self.ws_sender.start() + self.ws_sender_2.start() # 新增：启动第二个WebSocket服务 + self.processor.start() + for cam in self.cameras: + w = RTSPCaptureWorker(cam, self.raw_queue, self.stop_event) + w.start() + self.capture_workers.append(w) + print("[INFO] Kadian RTSP Service started") + + def stop(self): + self.stop_event.set() + self.raw_queue.join() + self.ws_queue.join() + self.ws_queue_2.join() # 新增：等待第二个WebSocket队列 + for w in self.capture_workers: + w.join(timeout=2.0) + self.processor.join(timeout=2.0) + self.ws_sender.join(timeout=2.0) + self.ws_sender_2.join(timeout=2.0) # 新增：等待第二个WebSocket发送器 + print("[INFO] Service stopped") + + +if __name__ == "__main__": + service = RTSPService("config.yaml") + service.start() + try: + while True: + time.sleep(1) + except KeyboardInterrupt: + service.stop()