昨天的晚会让人脸识别又火了，转载一篇来自 CSDN JireRen 的精彩博文，借花献佛给大家一起尝试DIY树莓派上的人脸识别。
使用树莓派2和OpenCV制作一个简易的人脸识别和追踪系统。
所需硬件
需要：树莓派2、Pi Camera
非必须（如果需要追踪人脸运动，需要一个有两个马达的小云台）：云台
安装OpenCV
sudo apt-get update sudo apt-get upgrade sudo apt-get install python-opencv
安装PiCamera
由于我没有使用USB摄像头，而是用了特殊的Pi Camera，样子如下图， 所以需要安装PiCamera来控制摄像头。
安装PiCamera:
sudo apt-get install python-pip sudo apt-get install python-dev sudo pip install picamera
至此人脸识别所需要的准备工作已经完成，可以使用下面的演示代码进行测试。
示例代码
Demo.1
第一个演示只使用单核，由于树莓派的性能有限，在只使用一个CPU核心的情况下视频的帧数非常之低，只有5帧左右，效果不太理想， 另外代码中通过Servo Blaster 控制云台的电机，来实现追踪人脸的功能，不过考虑到这个功能不是必须，所以不在此进行介绍。
### Imports ################################################################### from picamera.array import PiRGBArray from picamera import PiCamera import time import cv2 import os ### Setup ##################################################################### # Center coordinates cx = 160 cy = 120 os.system( “echo 0=150 > /dev/servoblaster” ) os.system( “echo 1=150 > /dev/servoblaster” ) xdeg = 150 ydeg = 150 # Setup the camera camera = PiCamera() camera.resolution = ( 320, 240 ) camera.framerate = 60 rawCapture = PiRGBArray( camera, size=( 320, 240 ) ) # Load a cascade file for detecting faces face_cascade = cv2.CascadeClassifier( ‘/home/pi/opencv-2.4.9/data/lbpcascades/lbpcascade_frontalface.xml’ ) t_start = time.time() fps = 0 ### Main ###################################################################### # Capture frames from the camera for frame in camera.capture_continuous( rawCapture, format=”bgr”, use_video_port=True ): image = frame.array # Use the cascade file we loaded to detect faces gray = cv2.cvtColor( image, cv2.COLOR_BGR2GRAY ) faces = face_cascade.detectMultiScale( gray ) print “Found ” + str( len( faces ) ) + ” face(s)” # Draw a rectangle around every face and move the motor towards the face for ( x, y, w, h ) in faces: cv2.rectangle( image, ( x, y ), ( x + w, y + h ), ( 100, 255, 100 ), 2 ) cv2.putText( image, “Face No.” + str( len( faces ) ), ( x, y ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) tx = x + w/2 ty = y + h/2 if ( cx – tx > 10 and xdeg <= 190 ): xdeg += 3 os.system( "echo 0=" + str( xdeg ) + " > /dev/servoblaster” ) elif ( cx – tx < -10 and xdeg >= 110 ): xdeg -= 3 os.system( “echo 0=” + str( xdeg ) + ” > /dev/servoblaster” ) if ( cy – ty > 10 and ydeg >= 110 ): ydeg -= 3 os.system( “echo 1=” + str( ydeg ) + ” > /dev/servoblaster” ) elif ( cy – ty < -10 and ydeg <= 190 ): ydeg += 3 os.system( "echo 1=" + str( ydeg ) + " > /dev/servoblaster” ) # Calculate and show the FPS fps = fps + 1 sfps = fps / ( time.time() – t_start ) cv2.putText( image, “FPS : ” + str( int( sfps ) ), ( 10, 10 ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) # Show the frame cv2.imshow( “Frame”, image ) cv2.waitKey( 1 ) # Clear the stream in preparation for the next frame rawCapture.truncate( 0 )
[WPGP gif_id=”3055″ width=”600″]
另外请注意由于我使用HaarCascade来进行人脸检测， 需要使用到识别人脸的XML，这些人脸识别的XML文件是随着OpenCV一起安装的，不需要额外的安装， 不过当你在自己树莓派上运行时，请注意调整XML文件的路径， 就是调整这一行：
# Load a cascade file for detecting faces face_cascade = cv2.CascadeClassifier( ‘你的XML文件路径’ )
Demo.2
通过同时使用不同的XML文件，可以实现同时识别不同物体的功能，比如下面这段代码可以同时识别人脸和黑色手机，识别手机所需要的XML文件是由Radamés Ajna和Thiago Hersan制作的， 来源在这里。 更进一步的，我们可以根据自己的需要训练自己的Cascade文件，Naotoshi Seo在此处 给出了详细的教程， 比较简易的还有Thorsten Ball的香蕉识别教程。
### Imports ################################################################### from picamera.array import PiRGBArray from picamera import PiCamera import time import cv2 import os import pygame ### Setup ##################################################################### os.putenv(‘SDL_FBDEV’, ‘/dev/fb1’) # Setup the camera camera = PiCamera() camera.resolution = ( 320, 240 ) camera.framerate = 40 rawCapture = PiRGBArray( camera, size=( 320, 240 ) ) # Load the cascade files for detecting faces and phones face_cascade = cv2.CascadeClassifier( ‘/home/pi/opencv-2.4.9/data/lbpcascades/lbpcascade_frontalface.xml’ ) phone_cascade = cv2.CascadeClassifier( ‘cascade.xml’ ) t_start = time.time() fps = 0 ### Main ###################################################################### # Capture frames from the camera for frame in camera.capture_continuous( rawCapture, format=”bgr”, use_video_port=True ): image = frame.array # Look for faces and phones in the image using the loaded cascade file gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY) faces = face_cascade.detectMultiScale(gray) phones = phone_cascade.detectMultiScale(gray) # Draw a rectangle around every face for (x,y,w,h) in faces: cv2.rectangle( image, ( x, y ), ( x + w, y + h ), ( 255, 255, 0 ), 2 ) cv2.putText( image, “Face No.” + str( len( faces ) ), ( x, y ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) # Draw a rectangle around every phone for (x,y,w,h) in phones: cv2.rectangle( image, ( x, y ), ( x + w, y + h ), ( 255, 0, 0 ), 2 ) cv2.putText( image, “iPhone”, ( x, y ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 255, 255 ), 2 ) # Calculate and show the FPS fps = fps + 1 sfps = fps / ( time.time() – t_start ) cv2.putText( image, “FPS : ” + str( int( sfps ) ), ( 10, 10 ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) cv2.imshow( “Frame”, image ) cv2.waitKey( 1 ) # Clear the stream in preparation for the next frame rawCapture.truncate( 0 )
[WPGP gif_id=”3053″ width=”600″]
由于使用了更多的XML文件进行识别，帧数降低到了2~3帧。
Demo.3
为了解决帧数较低的问题，有一个比较简单的方法就是跳帧，可以不对每一帧图像都进行识别，而是隔几帧识别一次（因为最初因为懒不想将程序写成多线程，但是为了提高帧数，所以有了这个蛋疼的方法…）。
### Imports ################################################################### from picamera.array import PiRGBArray from picamera import PiCamera import time import cv2 import os import pygame ### Setup ##################################################################### os.putenv( ‘SDL_FBDEV’, ‘/dev/fb1’ ) # Setup the camera camera = PiCamera() camera.resolution = ( 320, 240 ) camera.framerate = 30 rawCapture = PiRGBArray( camera, size=( 320, 240 ) ) fcounter = 0 facefind = 0 # Load a cascade file for detecting faces face_cascade = cv2.CascadeClassifier( ‘/home/pi/opencv-2.4.9/data/lbpcascades/lbpcascade_frontalface.xml’ ) t_start = time.time() fps = 0 ### Main ###################################################################### # Capture frames from the camera for frame in camera.capture_continuous( rawCapture, format=”bgr”, use_video_port=True ): image = frame.array # Run the face detection algorithm every four frames if fcounter == 3: fcounter = 0 # Look for faces in the image using the loaded cascade file gray = cv2.cvtColor( image, cv2.COLOR_BGR2GRAY ) faces = face_cascade.detectMultiScale( gray ) print “Found ” + str( len( faces ) ) + ” face(s)” if str( len( faces ) ) != 0: facefind = 1 facess = faces else: facefind = 0 # Draw a rectangle around every face for ( x, y, w, h ) in faces: cv2.rectangle( image, ( x, y ), ( x + w, y + h ), ( 200, 255, 0 ), 2 ) cv2.putText( image, “Face No.” + str( len( facess ) ), ( x, y ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) facess = faces else: if facefind == 1 and str( len( facess ) ) != 0: # Continue to draw the rectangle around every face for ( x, y, w, h ) in facess: cv2.rectangle( image, ( x, y ), ( x + w, y + h ), ( 200, 255, 0 ), 2 ) cv2.putText( image, “Face No.” + str( len( facess ) ), ( x, y ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) fcounter += 1 # Calculate and show the FPS fps = fps + 1 sfps = fps / ( time.time() – t_start ) cv2.putText( image, “FPS : ” + str( int( sfps ) ), ( 10, 10 ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) cv2.imshow( “Frame”, image ) cv2.waitKey( 1 ) # Clear the stream in preparation for the next frame rawCapture.truncate( 0 )
[WPGP gif_id=”3051″ width=”600″]
这样子帧数会提高到10帧左右，已经不像原来那么卡顿，但是当你移动速度很快的时候，识别框会出现滞后。
Demo.4
毕竟跳帧只是权宜之计，这个版本使用了全部的CPU核心，帧数稳定在了15帧左右。
### Imports ################################################################### from picamera.array import PiRGBArray from picamera import PiCamera from functools import partial import multiprocessing as mp import cv2 import os import time ### Setup ##################################################################### os.putenv( ‘SDL_FBDEV’, ‘/dev/fb0’ ) resX = 320 resY = 240 cx = resX / 2 cy = resY / 2 os.system( “echo 0=150 > /dev/servoblaster” ) os.system( “echo 1=150 > /dev/servoblaster” ) xdeg = 150 ydeg = 150 # Setup the camera camera = PiCamera() camera.resolution = ( resX, resY ) camera.framerate = 60 # Use this as our output rawCapture = PiRGBArray( camera, size=( resX, resY ) ) # The face cascade file to be used face_cascade = cv2.CascadeClassifier(‘/home/pi/opencv-2.4.9/data/lbpcascades/lbpcascade_frontalface.xml’) t_start = time.time() fps = 0 ### Helper Functions ########################################################## def get_faces( img ): gray = cv2.cvtColor( img, cv2.COLOR_BGR2GRAY ) faces = face_cascade.detectMultiScale( gray ) return faces, img def draw_frame( img, faces ): global xdeg global ydeg global fps global time_t # Draw a rectangle around every face for ( x, y, w, h ) in faces: cv2.rectangle( img, ( x, y ),( x + w, y + h ), ( 200, 255, 0 ), 2 ) cv2.putText(img, “Face No.” + str( len( faces ) ), ( x, y ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) tx = x + w/2 ty = y + h/2 if ( cx – tx > 15 and xdeg <= 190 ): xdeg += 1 os.system( "echo 0=" + str( xdeg ) + " > /dev/servoblaster” ) elif ( cx – tx < -15 and xdeg >= 110 ): xdeg -= 1 os.system( “echo 0=” + str( xdeg ) + ” > /dev/servoblaster” ) if ( cy – ty > 15 and ydeg >= 110 ): ydeg -= 1 os.system( “echo 1=” + str( ydeg ) + ” > /dev/servoblaster” ) elif ( cy – ty < -15 and ydeg <= 190 ): ydeg += 1 os.system( "echo 1=" + str( ydeg ) + " > /dev/servoblaster” ) # Calculate and show the FPS fps = fps + 1 sfps = fps / (time.time() – t_start) cv2.putText(img, “FPS : ” + str( int( sfps ) ), ( 10, 10 ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) cv2.imshow( “Frame”, img ) cv2.waitKey( 1 ) ### Main ###################################################################### if __name__ == ‘__main__’: pool = mp.Pool( processes=4 ) fcount = 0 camera.capture( rawCapture, format=”bgr” ) r1 = pool.apply_async( get_faces, [ rawCapture.array ] ) r2 = pool.apply_async( get_faces, [ rawCapture.array ] ) r3 = pool.apply_async( get_faces, [ rawCapture.array ] ) r4 = pool.apply_async( get_faces, [ rawCapture.array ] ) f1, i1 = r1.get() f2, i2 = r2.get() f3, i3 = r3.get() f4, i4 = r4.get() rawCapture.truncate( 0 ) for frame in camera.capture_continuous( rawCapture, format=”bgr”, use_video_port=True ): image = frame.array if fcount == 1: r1 = pool.apply_async( get_faces, [ image ] ) f2, i2 = r2.get() draw_frame( i2, f2 ) elif fcount == 2: r2 = pool.apply_async( get_faces, [ image ] ) f3, i3 = r3.get() draw_frame( i3, f3 ) elif fcount == 3: r3 = pool.apply_async( get_faces, [ image ] ) f4, i4 = r4.get() draw_frame( i4, f4 ) elif fcount == 4: r4 = pool.apply_async( get_faces, [ image ] ) f1, i1 = r1.get() draw_frame( i1, f1 ) fcount = 0 fcount += 1 rawCapture.truncate( 0 )
帧数上升到了13左右，而且识别框没有延迟。
[WPGP gif_id=”3048″ width=”600″]
Demo.5
搞定了低帧数问题，我又试了试多核加跳帧…帧数可到28帧左右。
### Imports ################################################################### from picamera.array import PiRGBArray from picamera import PiCamera from functools import partial import multiprocessing as mp import cv2 import os ### Setup ##################################################################### os.putenv( ‘SDL_FBDEV’, ‘/dev/fb0’ ) resX = 320 resY = 240 # Setup the camera camera = PiCamera() camera.resolution = ( resX, resY ) camera.framerate = 90 t_start = time.time() fps = 0 # Use this as our output rawCapture = PiRGBArray( camera, size=( resX, resY ) ) # The face cascade file to be used face_cascade = cv2.CascadeClassifier( ‘/home/pi/opencv-2.4.9/data/lbpcascades/lbpcascade_frontalface.xml’ ) ### Helper Functions ########################################################## def get_faces( img ): gray = cv2.cvtColor( img, cv2.COLOR_BGR2GRAY ) return face_cascade.detectMultiScale( gray ), img def draw_frame( img, faces ): global fps global time_t # Draw a rectangle around every face for ( x, y, w, h ) in faces: cv2.rectangle( img, ( x, y ),( x + w, y + h ), ( 200, 255, 0 ), 2 ) # Calculate and show the FPS fps = fps + 1 sfps = fps / (time.time() – t_start) cv2.putText(img, “FPS : ” + str( int( sfps ) ), ( 10, 10 ), cv2.FONT_HERSHEY_SIMPLEX, 0.5, ( 0, 0, 255 ), 2 ) cv2.imshow( “Frame”, img ) cv2.waitKey( 1 ) ### Main ###################################################################### if __name__ == ‘__main__’: pool = mp.Pool( processes=4 ) i = 0 rList = [None] * 17 fList = [None] * 17 iList = [None] * 17 camera.capture( rawCapture, format=”bgr” ) for x in range ( 17 ): rList[x] = pool.apply_async( get_faces, [ rawCapture.array ] ) fList[x], iList[x] = rList[x].get() fList[x] = [] rawCapture.truncate( 0 ) for frame in camera.capture_continuous( rawCapture, format=”bgr”, use_video_port=True ): image = frame.array if i == 1: rList[1] = pool.apply_async( get_faces, [ image ] ) draw_frame( iList[2], fList[1] ) elif i == 2: iList[2] = image draw_frame( iList[3], fList[1] ) elif i == 3: iList[3] = image draw_frame( iList[4], fList[1] ) elif i == 4: iList[4] = image fList[5], iList[5] = rList[5].get() draw_frame( iList[5], fList[5] ) elif i == 5: rList[5] = pool.apply_async( get_faces, [ image ] ) draw_frame( iList[6], fList[5] ) elif i == 6: iList[6] = image draw_frame( iList[7], fList[5] ) elif i == 7: iList[7] = image draw_frame( iList[8], fList[5] ) elif i == 8: iList[8] = image fList[9], iList[9] = rList[9].get() draw_frame( iList[9], fList[9] ) elif i == 9: rList[9] = pool.apply_async( get_faces, [ image ] ) draw_frame( iList[10], fList[9] ) elif i == 10: iList[10] = image draw_frame( iList[11], fList[9] ) elif i == 11: iList[11] = image draw_frame( iList[12], fList[9] ) elif i == 12: iList[12] = image fList[13], iList[13] = rList[13].get() draw_frame( iList[13], fList[13] ) elif i == 13: rList[13] = pool.apply_async( get_faces, [ image ] ) draw_frame( iList[14], fList[13] ) elif i == 14: iList[14] = image draw_frame( iList[15], fList[13] ) elif i == 15: iList[15] = image draw_frame( iList[16], fList[13] ) elif i == 16: iList[16] = image fList[1], iList[1] = rList[1].get() draw_frame( iList[1], fList[1] ) i = 0 i += 1 rawCapture.truncate( 0 )
[WPGP gif_id=”3047″ width=”600″]
跳帧加多核，强行30帧哈哈，不过还是建议最终使用Demo4。
这篇博客节选翻译自我自己的课程报告， 同样的内容也出现于我自己的英文博客， 最后出镜的是我的搭档Andre Heil。
via 转自 JireRen 的博客。