姿态估计 MediaPipe实现手势，人体姿态，面部动作估计的用法

在这里插入图片描述

import mediapipe as mp
import cv2
import numpy as np
import time
# 定义一个函数，计算两个点的距离
def findDis(pts1,pts2):
    return ((pts2[0]-pts1[0])**2 + (pts2[1]-pts1[1])**2)**0.5
# 创建手势检测模型
mpHands = mp.solutions.hands  # 检测人的手
hand_mode = mpHands.Hands(max_num_hands=2,min_detection_confidence=0.5,
                        min_tracking_confidence=0.5) 
# static_image_mode：默认为False，如果设置为false, 就是把输入看作一个视频流，在检测到手之后对手加了一个目标跟踪(目标检测+跟踪)，
# 无需调用另一次检测，直到失去对任何手的跟踪为止。如果设置为True，则手部检测将在每个输入图像上运行(目标检测)，非常适合处理一批静态的，
# 可能不相关的图像。(如果检测的是图片就要设置成True)
# 检测手的模式参数设置，max_num_hands:可以检测到的手的数量最大值，默认是2
# min_detection_confidence: 手部检测的最小置信度值，大于这个数值被认为是成功的检测，
# min_tracking_confidence：目标踪模型的最小置信度值，大于这个数值将被视为已成功跟踪的手部，如果static_image_mode设置为true，则忽略此操作。
mpDraw = mp.solutions.drawing_utils  # 绘图

cap = cv2.VideoCapture(0)

while True:
    success,img = cap.read()
    img = cv2.flip(img,1)
    results = hand_mode.process(img)# 将图片导入模型，获取20个点的坐标进行分析
    if results.multi_hand_landmarks:
        for handLms in results.multi_hand_landmarks:
            point4_8 = []
            for id,lm in enumerate(handLms.landmark):
                h,w,c = img.shape
                cx,cy = int(lm.x*w),int(lm.y*h)
                cv2.circle(img,(cx,cy),10,(255,0,0),-1)
                if id in [4,8]:# 获取点4，8的坐标
                    point4_8.append([cx,cy])
            mpDraw.draw_landmarks(img, handLms, mpHands.HAND_CONNECTIONS)
            # 求点4，8的坐标，进行可视化展示
            cv2.line(img,(point4_8[0][0],point4_8[0][1]),(point4_8[1][0],point4_8[1][1]),(0,0,255),5)
            distance = round(findDis((point4_8[0][0],point4_8[0][1]),(point4_8[1][0],point4_8[1][1])),2)
            cv2.putText(img,"distance:{}".format(distance),(50,50),cv2.FONT_HERSHEY_PLAIN, 3,(0,0,255),3)
            cv2.rectangle(img,(20,250),(20+10,250-int(distance)),(255,0,255),20)
    cv2.imshow("img",img)
    if cv2.waitKey(1)&0xFF == ord("q"):
        break
cap.release()
cv2.destroyAllWindows()

在这里插入图片描述
2. 人体姿势识别

mpPose = mp.solutions.pose  # 检测人的手
pose_mode = mpPose.Pose(min_detection_confidence=0.5,min_tracking_confidence=0.5)  # 模式参数设置

mpDraw = mp.solutions.drawing_utils  # 绘图

cap = cv2.VideoCapture(0)
biaoji = 0
i = 0

while True:
    success,img = cap.read()
    img = cv2.flip(img,1)
    results = pose_mode.process(img)
    if results.pose_landmarks:
        point23_25 = []
        for id,lm in enumerate(results.pose_landmarks.landmark):
            h,w,c = img.shape
            cx,cy = int(lm.x*w),int(lm.y*h)
            cv2.circle(img,(cx,cy),10,(255,0,0),-1)
            if id in [23,25]:
                point23_25.append([cx,cy])
        mpDraw.draw_landmarks(img, results.pose_landmarks, mpPose.POSE_CONNECTIONS)
        cv2.line(img,(point23_25[0][0],point23_25[0][1]),(point23_25[1][0],point23_25[1][1]),(0,0,255),5)
        if point23_25[0][1]>point23_25[1][1]:
            if biaoji == 1:
                i += 1
                biaoji = 0
                cv2.putText(img,"Leg up--{}".format(i),(10,50),cv2.FONT_HERSHEY_PLAIN, 3,(0,0,255),3)
        else:
            biaoji = 1
            cv2.putText(img,"Leg down--{}".format(i),(10,450),cv2.FONT_HERSHEY_PLAIN, 3,(0,0,255),3)
    cv2.imshow("img",img)
    if cv2.waitKey(1)&0xFF == ord("q"):
        break
cap.release()
cv2.destroyAllWindows()

在这里插入图片描述
3. 面部动作识别

def findDis(pts1,pts2):
    return ((pts2[0]-pts1[0])**2 + (pts2[1]-pts1[1])**2)**0.5

cap = cv2.VideoCapture(0)
pTime = 0

id_list = [23, 159, 130, 243, 62, 292, 12, 15]

mpDraw = mp.solutions.drawing_utils
mpFaceMesh = mp.solutions.face_mesh
faceMesh = mpFaceMesh.FaceMesh(max_num_faces=2)
drawSpec = mpDraw.DrawingSpec(thickness=1, circle_radius=2)

while True:
    success, img = cap.read()
    imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    results = faceMesh.process(imgRGB)
    if results.multi_face_landmarks:
        for faceLms in results.multi_face_landmarks:
            mpDraw.draw_landmarks(img, faceLms, mpFaceMesh.FACEMESH_CONTOURS,
                                  drawSpec,drawSpec)
            mp_data = []
            for id,lm in enumerate(faceLms.landmark):
                ih, iw, ic = img.shape
                x,y = int(lm.x*iw), int(lm.y*ih)
                if id in id_list:           #左眼[22, 23, 24, 26, 110, 157, 158, 159, 160, 161, 130, 243]:
                    mp_data.append([x,y])
                    cv2.circle(img,(x,y),2,(255,0,0),-1)
            eye_length_1 = findDis(mp_data[0],mp_data[1])
            eye_length_2 = findDis(mp_data[2],mp_data[3])
            mouth_length_2 = findDis(mp_data[4],mp_data[5])
            mouth_length_1 = findDis(mp_data[6],mp_data[7])
#             print(eye_length_1,eye_length_2)
            if ((mouth_length_1/mouth_length_2)<(98/18)):
                cv2.putText(img,"mouth close",(400,50),cv2.FONT_HERSHEY_PLAIN, 2,(0,0,255),3)
                    
            else:
                cv2.putText(img,"mouth open",(400,50),cv2.FONT_HERSHEY_PLAIN, 2,(0,0,255),3)
            if (eye_length_2/eye_length_1)>18:
                cv2.putText(img,"eye open",(400,100),cv2.FONT_HERSHEY_PLAIN, 2,(0,0,255),3)  
            else:                 
                cv2.putText(img,"eye close",(400,100),cv2.FONT_HERSHEY_PLAIN, 2,(0,0,255),3)  

    cTime = time.time()
    fps = 1 / (cTime - pTime)
    pTime = cTime
    cv2.putText(img, f'FPS: {int(fps)}', (20, 70), cv2.FONT_HERSHEY_PLAIN,
                3, (255, 0, 0), 3)
    cv2.imshow("Image", img)
    if cv2.waitKey(1)&0xFF == ord("q"):
        cv2.imwrite("6.jpg",img)
        break
cap.release()
cv2.destroyAllWindows()