基于dlib的抽烟检测

detector = dlib.get_frontal_face_detector()
    predictor = dlib.shape_predictor("./shape_predictor_68_face_landmarks.dat")

    
    cap = cv2.VideoCapture("smoke.avi")
    ret, image = cap.read()
    idxx=1
    while ret:
        print("id",idxx)
        # sta = time.time()
        ret, image = cap.read()

      
        # image = cv2.resize(image, (640,384))
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

        rects = detector(gray, 1)

        # enumerate()方法用于将一个可遍历的数据对象(列表、元组、字典)组合
        # 为一个索引序列，同时列出 数据下标 和 数据 ，一般用在for循环中
        check_smoke=False
        if rects is None:
            continue
        for (i, rect) in enumerate(rects):
            shape = predictor(gray, rect)  # 标记人脸中的68个landmark点
            shape = face_utils.shape_to_np(shape)  # shape转换成68个坐标点矩阵
            (mStart, mEnd) = face_utils.FACIAL_LANDMARKS_68_IDXS["mouth"]

            (x_face, y_face, w_face, h_face) = face_utils.rect_to_bb(rect)  # 返回人脸框的左上角坐标和矩形框的尺寸
            # cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 2)

            # cv2.putText(image, "Face #{}".format(i + 1), (x - 10, y - 10),
            #             cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)

            # landmarksNum = 0;
            m_49 = None
            m_55 =