real_gesture_copy.py

import cv2
import STESDK
import numpy as np
from STESDK import HandDetector, HandAligner, HandTracker, BodyDetector, BodyAligner
from yolov7.detect_phone import load_yolo_model, yolo_detect_frame, rec_phone

import os
import sys

# 将项目根目录添加到sys.path
project_root = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, project_root)

weights_path = 'yolov7/weights/result_050702.pt'
img_size = 640
device = '0'  # 使用GPU设备编号，如果使用CPU，请设置为 'cpu'

# 初始化YOLOv7模型
model, device, half = load_yolo_model(weights_path, img_size, device)


# 定义函数判断是否为打电话姿态
def is_phone_call_gesture(face_points, hand_avg_point, threshold):
    # 计算手部关键点平均位置与人脸关键点的距离
    distances = [np.sqrt((point[0] - hand_avg_point[0]) ** 2 + (point[1] - hand_avg_point[1]) ** 2) for point in
                 face_points]
    # 如果距离小于阈值，则判断为打电话姿态
    return any(distance < threshold for distance in distances)


# 初始化手部检测器、对齐器和跟踪器
hand_detector = HandDetector()
hand_aligner = HandAligner()
hand_tracker = HandTracker()

# 初始化人体检测器和对齐器
body_detector = BodyDetector()
body_aligner = BodyAligner()

# 打开摄像头
cap = cv2.VideoCapture(0)

# 设置分辨率
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)

# 设置距离阈值，您可以根据实际情况调整这个值
threshold = 150

# 检测间隔（以帧为单位）
detection_interval = 5

frame_counter = 0

conf_thres = 0.70  # 检测的置信度阈值，您可以根据需要调整
iou_thres = 0.45  # NMS的IoU阈值，您可以根据需要调整
augment = False  # 是否进行数据增强
classes = None  # 检测的类别，设置为None表示检测所有类别
agnostic_nms = False  # 如果为True，则在NMS中对类别进行归一化处理

while True:
    ret, frame = cap.read()
    if ret:
        frame_counter += 1

        if frame_counter % detection_interval == 0:
            # 跟踪手部并渲染边框
            rects = hand_tracker.track(frame)

            # 如果检测到手
            if rects:
                # 获取手部关键点
                hand_points = hand_aligner.align(frame, rects[0])

                # 计算关键点的平均位置
                if len(hand_points) > 0:
                    avg_x = sum(point[0] for point in hand_points) / len(hand_points)
                    avg_y = sum(point[1] for point in hand_points) / len(hand_points)
                    hand_avg_point = (int(avg_x), int(avg_y))

                    # 在帧上绘制平均位置点
                    cv2.circle(frame, hand_avg_point, 5, (0, 255, 0), -1)

                    # 输出平均位置
                    print(f"Average position of hand keypoints: {hand_avg_point}")

                    # 人体位置检测
                    body_rect = body_detector.detect(frame)

                    # 如果检测到人体
                    if body_rect:
                        # 关键点检测
                        body_points = body_aligner.align(frame, body_rect[0])

                        # 获取两个新的关键点（人脸关键点的两个位置信息）
                        y = (body_points[0][1] + body_points[1][1]) / 2
                        x2 = (body_points[1][0] + body_points[3][0]) / 2
                        x1 = (body_points[1][0] + body_points[2][0]) / 2
                        new_points = [(int(x1), int(y)), (int(x2), int(y))]

                        # 判断是否为打电话姿态
                        if is_phone_call_gesture(new_points, hand_avg_point, threshold) and yolo_detect_frame(model,device, half,frame,img_size, conf_thres,iou_thres,augment,classes,agnostic_nms):
                            detection_time, label, x1, x2, y1, y2 = rec_phone(model, device, half, frame, img_size,
                                                                              conf_thres, iou_thres, augment, classes,
                                                                              agnostic_nms)
                            print(
                                f"帧数：{frame_counter},推理时延: {detection_time:.2f}ms, 标签: ({label}), 预测框坐标: ({x1:.2f}), ({x2:.2f}) , ({y1:.2f}), {y2:.2f}),安全设备：商汤科技")

                        else:
                            detection_time = rec_phone(model, device, half, frame, img_size, conf_thres, iou_thres,
                                                       augment, classes, agnostic_nms)
                            print(f"帧数：{frame_counter}，推理时延: {detection_time:}ms")

        # 显示帧
        STESDK.imshow("hand keypoints with average", frame)

        # 等待按键
        k = cv2.waitKey(30)
        # 如果按下 'q' 键，退出循环
        if k == ord("q"):
            cv2.destroyAllWindows()
            break
    else:
        print("frame is empty")