树莓派4opencv实时人脸识别

import multiprocessing as mp
import cv2
import dlib

def detect_face(frame,value):
    img = frame
    dets = detector(img, 1)
    for index, face in enumerate(dets):
 
        left = face.left()
        top = face.top()
        right = face.right()
        bottom = face.bottom()

#因为如果没有检测到脸，上面四个变量是不存在的，
#会报出UnboundLocalError的错误，所以要处理一下。           
    try:
        value[:] = [left,top,right,bottom]
    except UnboundLocalError:
        value[:] = [0,0,0,0]

def draw_line(img,box):
    
    left = box[0]
    top = box[1]
    right = box[2]
    bottom = box[3]
    
#给传进来的img画框，并返回
    cv2.rectangle(img, (left*2, top*2), (right*2, bottom*2), (255, 0, 0), 1)
    return img
    
if __name__=='__main__':

#initial detector and cap
#树莓派的运行内存有限，所以要将采集的图片像素缩小一些，便于计算。
    detector = dlib.get_frontal_face_detector() #获取人脸分类器for face detection
    cap = cv2.VideoCapture(-1)  #Turn on the camera
    cap.set(cv2.CAP_PROP_FOURCC, cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'))
    cap.set(3,320)
    cap.set(4,240)

#initial boxes
#初始化框脸的初始位置
    box1 = mp.Array('i',[0,0,0,0])
    box2 = mp.Array('i',[0,0,0,0])
    
#initial Windowbox
    cv2.namedWindow('success', cv2.WINDOW_AUTOSIZE)
    
#initial frames and processes
#想要几个进程处理图片就用几组，但并不是用的越多越好，
#树莓派的CPU一共有4个核，全部用上可能会影响其他的性能，自己试的时候2个会好一点。
    ret, frame11 = cap.read()
    img11 = cv2.resize(frame11,(160,120))
    res1 = mp.Process(target=detect_face,args=(img11,box1))
    res1.start()

#可以把识别用的图片像素缩小，可以加快速度，同时也可以减少cpu负担，
#然后再把识别框扩大相应倍数，在原图片上放出。
    ret, frame21 = cap.read()
    img21 = cv2.resize(frame21,(160,120))
    res2 = mp.Process(target=detect_face,args=(img21,box2))
    res2.start()
    
    while(cap.isOpened()):
#process 1
 #如果想要逐帧处理，那就用pass，如果想跳帧就选择上面两句
        if (res1.is_alive()):
            ret, frame12 = cap.read()
            frame12 = cv2.flip(frame, 1)
            cv2.imshow('success',draw_line(frame12,box1))            
#            pass           
        else:
            ret, frame11 = cap.read()
            frame11 = cv2.flip(frame, 1)
            cv2.imshow('success',draw_line(frame11,box1))
          	img11 = cv2.resize(frame11,(160,120))           
            res1 = mp.Process(target=detect_face,args=(img11,box1))
            res1.start()
            
#process 2
        if (res2.is_alive()):
            ret, frame22 = cap.read()
            frame22 = cv2.flip(frame, 1)
            cv2.imshow('success',draw_line(frame22,box2))  
#            pass          
        else:
            ret, frame21 = cap.read()
            frame21 = cv2.flip(frame, 1)
            cv2.imshow('success',draw_line(frame21,box2))
          	img21 = cv2.resize(frame21,(160,120)) 
            res2 = mp.Process(target=detect_face,args=(img21,box2))
            res2.start()

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
            
    cv2.destroyAllWindows()
    cap.release()
    print('END')