python-OpenCV 视频中的手部跟踪： 基于mediapipe库

 直接看代码：

import cv2
import math
import time
import mediapipe as mp
from os import listdir

mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands

#调用mediapipe库中的Hand函数，这个函数就可以识别手部
hands = mp_hands.Hands(min_detection_confidence=0.5, min_tracking_confidence=0.5)

cap = cv2.VideoCapture(0)
while cap.isOpened():
    success, image = cap.read()
    if not success:
        print("没有图像")
        continue

    image = cv2.cvtColor(cv2.flip(image, 1), cv2.COLOR_BGR2RGB)
    image.flags.writeable = False
    #result是调用hands函数得到的结果，里面包含了手的各点坐标，通过results.multi_hand_landmarks调用
    results = hands.process(image)


    image.flags.writeable = True
    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
    if results.multi_hand_landmarks:
        #得到results.multi_hand_landmarks中所有点的坐标，并进行绘制
        for hand_landmarks in results.multi_hand_landmarks:
            mp_drawing.draw_landmarks(image, hand_landmarks, mp_hands.HAND_CONNECTIONS)


    cv2.imshow('result', image)
    if cv2.waitKey(5) & 0xFF == 27:
        break
cv2.destroyAllWindows()
hands.close()
cap.release()

需要注意的是，results.multi_hand_landmarks中的坐标一共有21个，对应了手的21个不同“关节”：

 结果：