基于 Mediapipe 手部识别的动作检测

使用 Google 的 Mediapipe 解决方案来检测手部的“OK”手势。以及根据拇指和食指的位置来确定当前检测到的手是左手还是右手。

环境准备

• 安装 OpenCV: pip install opencv-python
• 安装 Mediapipe: pip install mediapipe

代码

import cv2
import mediapipe as mp
import numpy as np

mp_hands = mp.solutions.hands # 加载手部关键点模型
mp_drawing = mp.solutions.drawing_utils # 加载绘图工具


def is_ok_gesture(landmarks): # 判断是否是 OK 手势
    thumb_tip = np.array(
        [landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP].x, landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP].y]) # 获取大拇指尖的坐标
    index_finger_tip = np.array([landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP].x,
                                 landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP].y])  # 获取食指尖的坐标

    # 计算大拇指和食指尖的距离
    distance = np.linalg.norm(thumb_tip - index_finger_tip)

    # 检查其它三个手指是否伸直
    middle_finger_tip = landmarks.landmark[mp_hands.HandLandmark.MIDDLE_FINGER_TIP].y
    ring_finger_tip = landmarks.landmark[mp_hands.HandLandmark.RING_FINGER_TIP].y
    pinky_tip = landmarks.landmark[mp_hands.HandLandmark.PINKY_TIP].y

    if distance < 0.05 and middle_finger_tip < ring_finger_tip and ring_finger_tip < pinky_tip: 
        return True
    return False

def detect_hand_actions(image):
    with mp_hands.Hands(min_detection_confidence=0.2, min_tracking_confidence=0.5) as hands: # 设置置信度
        image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) 
        results = hands.process(image_rgb) # 处理图像
        if results.multi_hand_landmarks: # 检查是否检测到手部
            for landmarks in results.multi_hand_landmarks: # 遍历每一只手
                mp_drawing.draw_landmarks(image, landmarks, mp_hands.HAND_CONNECTIONS) # 绘制关键点连线

                if is_ok_gesture(landmarks): # 判断是否是 OK 手势
                    cv2.putText(image, 'OK 手势', (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2,
                                cv2.LINE_AA)

                # 以下代码是获取手指的关键点
                thumb_tip = landmarks.landmark[mp_hands.HandLandmark.THUMB_TIP]
                index_finger_tip = landmarks.landmark[mp_hands.HandLandmark.INDEX_FINGER_TIP]

                # 这里只是一个简单示例，您可以添加更多的逻辑来识别特定的手部动作
                if thumb_tip.x > index_finger_tip.x:
                    print("右手")
                else:
                    print("左手")



    return image


cap = cv2.VideoCapture(0)
while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        continue

    image = detect_hand_actions(frame)
    cv2.imshow('Hand Action Detection', image)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()