也许是手势识别最简单的实现

import cv2
import mediapipe as mp
import cmath
import time
import numpy as np
import math
gesture = ["none","one","two","three","four","five","six","seven","eight","nine","ten"]
flag = 0
pTime = 0
mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands
mp_drawing_styles = mp.solutions.drawing_styles

hands = mp_hands.Hands(
        static_image_mode=False,
        max_num_hands=2,
        min_detection_confidence=0.5,
        min_tracking_confidence=0.75)

cap = cv2.VideoCapture(0)

# 记录中指轨迹的变量
prev_x, prev_y = None, None

# 创建画板，黑色背景
canvas = np.zeros((480, 640, 3), dtype=np.uint8)

# 获取画面宽度、高度
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))

rect=False
# 方块初始数组
x = 100
y = 100
w = 100
h = 100

L1 = 0
L2 = 0
on_square = False
square_color = (0, 255, 0)
while True:
    flag = 0
    ret,frame = cap.read()
    frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
    # 因为摄像头是镜像的，所以将摄像头水平翻转
    # 不是镜像的可以不翻转
    frame= cv2.flip(frame,1)
    results = hands.process(frame)
    frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
    
#    if results.multi_handedness:
#        for hand_label in results.multi_handedness:
#            print(hand_label)
            
    if results.multi_hand_landmarks:
        for hand_landmarks in results.multi_hand_landmarks:
            #print('hand_landmarks:', hand_landmarks)
            #计算关键点的距离，用于判断手指是否伸直
            p0_x = hand_landmarks.landmark[0].x
            p0_y = hand_landmarks.landmark[0].y
            p5_x = hand_landmarks.landmark[5].x
            p5_y = hand_landmarks.landmark[5].y
            distance_0_5 = pow(p0_x-p5_x,2)+pow(p0_y-p5_y,2)
            base = distance_0_5 / 0.6
            
            p4_x = hand_landmarks.landmark[4].x
            p4_y = hand_landmarks.landmark[4].y
            distance_5_4 = pow(p5_x-p4_x,2)+pow(p5_y-p4_y,2)
            
            p8_x = hand_landmarks.landmark[8].x
            p8_y = hand_landmarks.landmark[8].y
            distance_0_8 = pow(p0_x-p8_x,2)+pow(p0_y-p8_y,2)
            
            p12_x = hand_landmarks.landmark[12].x
            p12_y = hand_landmarks.landmark[12].y
            distance_0_12 = pow(p0_x-p12_x,2)+pow(p0_y-p12_y,2)
            
            p16_x = hand_landmarks.landmark[16].x
            p16_y = hand_landmarks.landmark[16].y
            distance_0_16 = pow(p0_x-p16_x,2)+pow(p0_y-p16_y,2)
            
            p20_x = hand_landmarks.landmark[20].x
            p20_y = hand_landmarks.landmark[20].y
            distance_0_20 = pow(p0_x-p20_x,2)+pow(p0_y-p20_y,2)
  
            if distance_0_8 > base:
                flag += 1
            if distance_0_12 > base:
                flag += 1
            if distance_0_16 > base:
                flag += 1
            if distance_0_20 > base:
                flag += 1
            if distance_5_4 > base*0.2:
                flag += 1
            if flag>=10:
                flag = 10
            if flag==5:
                # 获取中指关键点的坐标
                middle_finger = hand_landmarks.landmark[mp_hands.HandLandmark.MIDDLE_FINGER_TIP]
                x = int(middle_finger.x * frame.shape[1])
                y = int(middle_finger.y * frame.shape[0])
                
                # 绘制中指指点
                cv2.circle(frame, (x, y), 5, (0, 0, 255), -1)

                # 绘制中指轨迹
                if prev_x is not None and prev_y is not None:
                    cv2.line(canvas, (prev_x, prev_y), (x, y), (0, 255, 0), 3)
                
                # 更新前一个中指指点的位置
                prev_x, prev_y = x, y
            if flag==3:
                canvas = np.zeros((480, 640, 3), dtype=np.uint8)
                prev_x, prev_y = None, None
            if flag==2:

                rect=True

                x_list = []
                y_list = []
                for landmark in hand_landmarks.landmark:
                    x_list.append(landmark.x)
                    y_list.append(landmark.y)
                    
                
                # 获取食指指尖
                index_finger_x, index_finger_y= int(x_list[8] * width),int(y_list[8] * height)
    
                # 获取中指
                middle_finger_x,middle_finger_y =int(x_list[12] * width), int(y_list[12] * height)
                # 计算两指尖距离
                finger_distance =math.hypot((middle_finger_x - index_finger_x), (middle_finger_y - index_finger_y))
    
                # 如果双指合并(两之间距离近)
                if finger_distance < 60:
    
                    # X坐标范围 Y坐标范围
                    if (index_finger_x > x and index_finger_x < (x + w)) and (
                            index_finger_y > y and index_finger_y < (y + h)):
    
                        if on_square == False:
                            L1 = index_finger_x - x
                            L2 = index_finger_y - y
                            square_color = (255, 0, 255)
                            on_square = True
    
                else:
                    # 双指不合并/分开
                    on_square = False
                    square_color = (0, 255, 0)
    
                # 更新坐标
                if on_square:
                    x=index_finger_x - L1
                    y =index_finger_y - L2
                    # 图像融合 使方块不遮挡视频图片

            if flag!=2:
                rect=False

            # 关键点可视化
            mp_drawing.draw_landmarks(
                frame,
                hand_landmarks,
                mp_hands.HAND_CONNECTIONS,
                mp_drawing_styles.get_default_hand_landmarks_style(),
                mp_drawing_styles.get_default_hand_connections_style())


    if rect:
        cv2.rectangle(frame, (x, y), (x + w, y + h), square_color, -1)
        

    # 将画板叠加到原始图像上
    frame = cv2.addWeighted(frame, 1, canvas, 0.5, 0)
    cv2.putText(frame,gesture[flag],(50,50),0,1.3,(0,0,255),3)
    cTime = time.time()
    fps = 1 / (cTime - pTime)
    pTime = cTime
    cv2.putText(frame, f'FPS: {int(fps)}', (480, 60), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 124, 90), 1)
    cv2.imshow('MediaPipe', frame)
    if cv2.waitKey(1) & 0xFF == 13:
        break
    
cap.release()
cv2.destroyAllWindows()