MediaPipe人体姿态、手指关键点检测

MediaPipe人体姿态、手指关键点检测

---

前言

Mediapipe是google的一个开源项目，用于构建机器学习管道

提供了16个预训练模型的案例：人脸检测、Face Mesh、虹膜、手、姿态、人体、人物分割、头发分割、目标检测、Box Tracking、Instant Motion Tracking、3D目标检测、特征匹配、AutoFlip、MediaSequence、YouTube-8M

肢体识别本质上还是分类任务，该技术有很多应用场景，比如手势识别控制类应用、动作检测类应用、动作评测类应用、以及一些移动设备AR视频合成类应用

---

一、手指关键点检测

检测一只手中21个关键点坐标，每个点对应一个名称

import cv2
import mediapipe as mp
# 用来在图片中画出关键点
mp_drawing = mp.solutions.drawing_utils
# 关键点图样式
mp_drawing_styles = mp.solutions.drawing_styles
mp_hands = mp.solutions.hands
#MAX_NUM_HANDS 要检测的最大手数 默认为2
hands = mp_hands.Hands(max_num_hands=1)

if __name__ == '__main__':
    file = '1.png'
    # 图片翻转
    image = cv2.flip(cv2.imread(file), 1)
    results = hands.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
    image_height, image_width, _ = image.shape
    annotated_image = image.copy()
    # 遍历检测出来的手掌
    for hand_landmarks in results.multi_hand_landmarks:
        for lm in hand_landmarks.landmark:
            # 获取每个点的坐标
            x = lm.x * image_width
            y = lm.y * image_height

        # 画关键点
        mp_drawing.draw_landmarks(
            annotated_image,
            hand_landmarks,
            mp_hands.HAND_CONNECTIONS,
            mp_drawing_styles.get_default_hand_landmarks_style(),
            mp_drawing_styles.get_default_hand_connections_style())

    cv2.imwrite('0.png', cv2.flip(annotated_image, 1))

二、姿态检测

0bc36qaayaaa6qaoryyfmfrvb5gdb

import cv2
import mediapipe as mp
# 用来在图片中画出关键点
mp_drawing = mp.solutions.drawing_utils
# 关键点图样式
mp_drawing_styles = mp.solutions.drawing_styles
mpPose = mp.solutions.pose
pose = mpPose.Pose()

if __name__ == '__main__':

    file = '1.png'
    image = cv2.imread(file)
    image_height, image_width, _ = image.shape

    imgRGB = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    results = pose.process(imgRGB)
    # mediapipe姿态检测只能检测一个人
    if results.pose_landmarks:
        for lm in results.pose_landmarks.landmark:
            h, w, c = image.shape
            cx, cy = int(lm.x * w), int(lm.y * h)

    # 画关键点
    annotated_image = image.copy()
    mp_drawing.draw_landmarks(
        annotated_image,
        results.pose_landmarks,
        mpPose.POSE_CONNECTIONS,
        landmark_drawing_spec=mp_drawing_styles.get_default_pose_landmarks_style())
    cv2.imwrite('0.png', annotated_image)

三、3D物体案例检测案例

其他案例具体可以查看官网详细介绍：

https://google.github.io/mediapipe/solutions/pose.html