0x00. 平均值法

通过计算两帧图像之间变化了的像素点占的百分比，来确定图像中是否有动作产生。

这里主要用到 Absdiff 函数，比较两帧图像之间有差异的点，当然需要将图像进行一些处理，例如平滑处理，灰度化处理，二值化处理，经过处理之后的二值图像上的点将更有效。

代码示例：

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849 import cv2.cv as cv capture=cv.CaptureFromCAM(0) frame1 = cv.QueryFrame(capture)frame1gray = cv.CreateMat(frame1.height, frame1.width, cv.CV_8U)cv.CvtColor(frame1, frame1gray, cv.CV_RGB2GRAY) res = cv.CreateMat(frame1.height, frame1.width, cv.CV_8U) frame2gray = cv.CreateMat(frame1.height, frame1.width, cv.CV_8U) w= frame2gray.widthh= frame2gray.heightnb_pixels = frame2gray.width * frame2gray.height while True:    frame2 = cv.QueryFrame(capture)    cv.CvtColor(frame2, frame2gray, cv.CV_RGB2GRAY)     cv.AbsDiff(frame1gray, frame2gray, res)    cv.ShowImage(\”After AbsDiff\”, res)     cv.Smooth(res, res, cv.CV_BLUR, 5,5)    element = cv.CreateStructuringElementEx(5*2+1, 5*2+1, 5, 5,  cv.CV_SHAPE_RECT)    cv.MorphologyEx(res, res, None, None, cv.CV_MOP_OPEN)    cv.MorphologyEx(res, res, None, None, cv.CV_MOP_CLOSE)    cv.Threshold(res, res, 10, 255, cv.CV_THRESH_BINARY_INV)     cv.ShowImage(\”Image\”, frame2)    cv.ShowImage(\”Res\”, res)     #———–    nb=0    for y in range(h):        for x in range(w):            if res[y,x] == 0.0:                nb += 1    avg = (nb*100.0)/nb_pixels    #print \”Average: \”,avg, \”%r\”,    if avg >= 5:        print \”Something is moving !\”    #———–      cv.Copy(frame2gray, frame1gray)    c=cv.WaitKey(1)    if c==27: #Break if user enters \’Esc\’.        break

0x01. 背景建模与前景检测

背景建模也是检测运动物体的一种办法，下面是代码示例：

1234567891011121314151617181920212223242526272829303132333435363738 import cv2.cv as cv capture = cv.CaptureFromCAM(0)width = int(cv.GetCaptureProperty(capture, cv.CV_CAP_PROP_FRAME_WIDTH))height = int(cv.GetCaptureProperty(capture, cv.CV_CAP_PROP_FRAME_HEIGHT)) gray = cv.CreateImage((width,height), cv.IPL_DEPTH_8U, 1) background = cv.CreateMat(height, width, cv.CV_32F)backImage = cv.CreateImage((width,height), cv.IPL_DEPTH_8U, 1)foreground = cv.CreateImage((width,height), cv.IPL_DEPTH_8U, 1)output = cv.CreateImage((width,height), 8, 1) begin = Truethreshold = 10 while True:  frame = cv.QueryFrame( capture )   cv.CvtColor(frame, gray, cv.CV_BGR2GRAY)   if begin:      cv.Convert(gray, background) #Convert gray into background format      begin = False   cv.Convert(background, backImage) #convert existing background to backImage   cv.AbsDiff(backImage, gray, foreground) #Absdiff to get differences   cv.Threshold(foreground, output, threshold, 255, cv.CV_THRESH_BINARY_INV)   cv.Acc(foreground, background,output) #Accumulate to background   cv.ShowImage(\”Output\”, output)  cv.ShowImage(\”Gray\”, gray)  c=cv.WaitKey(1)  if c==27: #Break if user enters \’Esc\’.    break

0x02. 我的方法

上面的几种办法我都试了下，基本上能识别出运动的物体，但是发现总是有点瑕疵，所以又比对了几种别人的方案，然后合成了一个自己的方案：

具体处理思路：

  • 对两帧图像做一个absdiff得到新图像。
  • 对新图像做灰度和二值化处理。
  • 使用findContours函数获取二值化处理之后的图片中的轮廓。
  • 使用contourArea()过滤掉自己不想要的面积范围的轮廓。

这个办法基本上能够检测出物体的图像中物体的移动，而且我觉得通过设定contourArea()函数的过滤范围，可以检测距离摄像头不同距离范围的运动物体。

以下是代码示例：

1234567891011121314151617181920212223242526272829303132333435363738394041 #!usr/bin/env python#coding=utf-8 import cv2import numpy as np camera = cv2.VideoCapture(0)width = int(camera.get(3))height = int(camera.get(4)) firstFrame = None while True:  (grabbed, frame) = camera.read()  gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)  gray = cv2.GaussianBlur(gray, (21, 21), 0)   if firstFrame is None:    firstFrame = gray    continue   frameDelta = cv2.absdiff(firstFrame, gray)  thresh = cv2.threshold(frameDelta, 25, 255, cv2.THRESH_BINARY)[1]  # thresh = cv2.adaptiveThreshold(frameDelta,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,\\            # cv2.THRESH_BINARY,11,2)  # thresh = cv2.adaptiveThreshold(frameDelta,255,cv2.ADAPTIVE_THRESH_MEAN_C,\\  #           cv2.THRESH_BINARY,11,2)  thresh = cv2.dilate(thresh, None, iterations=2)  (_, cnts, _) = cv2.findContours(thresh.copy(), cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)  for c in cnts:     if cv2.contourArea(c) < 10000:       continue     (x, y, w, h) = cv2.boundingRect(c)      cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 2)    cv2.imshow(\”Security Feed\”, frame)    firstFrame = gray.copy()camera.release()cv2.destroyAllWindows()