利用光流法计算人体运动的速度与方向

利用光流法计算人体运动的速度与方向

1.方向的计算
首先计算图像各个象素的光流(opencv LK),然后建立4*4窗口对X,Y方向分别做统计求和,
然后求得 atan(yy/xx)作为光流方向,即为运动的方向.
 

2.速度的计算
利用帧差分得到运动图像,然后建立4*4窗口对图像进行统计求和,求和值作为权重,表示速度的比例.
即运动区域白色(255)面积越大,速度越大.

3.结果
大部分运动方向计算正确,少部分有错误,还需要改进算法.(利用统计?)

 

4.代码:

WW_RETURN HumanMotion::ImgOpticalFlow(IplImage *pre_grey,IplImage *grey)
/*************************************************
  Function:
  Description:  光流法计算运动速度与方向      
  Date:   2006－6－14
  Author:  
  Input:                       
  Output:        
  Return:        
  Others:         
*************************************************/
{

 IplImage *velx = cvCreateImage( cvSize(grey->width ,grey->height),IPL_DEPTH_32F, 1 );
 IplImage *vely = cvCreateImage( cvSize(grey->width ,grey->height),IPL_DEPTH_32F, 1 );

 velx->origin =  vely->origin = grey->origin;
 CvSize winSize = cvSize(5,5);
 cvCalcOpticalFlowLK( prev_grey, grey, winSize, velx, vely );
 
 cvAbsDiff( grey,prev_grey, abs_img );
 cvThreshold( abs_img, abs_img, 29, 255, CV_THRESH_BINARY); 

 CvScalar xc,yc; 
 for(int y =0 ;y<velx->height; y++)
  for(int x =0;x<velx->width;x++ )
  {
   xc = cvGetAt(velx,y,x);
   yc = cvGetAt(vely,y,x);

   
   float x_shift= (float)xc.val[0];
   float y_shift= (float)yc.val[0];
   const int winsize=5;  //计算光流的窗口大小

   if((x%(winsize*2)==0) && (y%(winsize*2)==0) )
   {

    if(x_shift!=0 || y_shift!=0)
    {
     
     if(x>winsize && y>winsize && x <(velx->width-winsize) && y<(velx->height-winsize) )
     {

      cvSetImageROI( velx, cvRect( x-winsize, y-winsize, 2*winsize, 2*winsize));
      CvScalar total_x = cvSum(velx);
      float xx = (float)total_x.val[0];
      cvResetImageROI(velx);

      cvSetImageROI( vely, cvRect( x-winsize, y-winsize, 2*winsize, 2*winsize));
      CvScalar total_y = cvSum(vely);
      float yy = (float)total_y.val[0];
      cvResetImageROI(vely);
      
      cvSetImageROI( abs_img, cvRect( x-winsize, y-winsize, 2*winsize, 2*winsize));
      CvScalar total_speed = cvSum(abs_img);
      float ss = (float)total_speed.val[0]/(4*winsize*winsize)/255;
      cvResetImageROI(abs_img);

      const double ZERO = 0.000001;
      const double pi = 3.1415926;
      double alpha_angle;

      if(xx<ZERO && xx>-ZERO)
       alpha_angle = pi/2;
      else
       alpha_angle = abs(atan(yy/xx));
      
      if(xx<0 && yy>0) alpha_angle = pi - alpha_angle ;
      if(xx<0 && yy<0) alpha_angle = pi + alpha_angle ;
      if(xx>0 && yy<0) alpha_angle = 2*pi - alpha_angle ;

      
      CvScalar line_color;
      float scale_factor = ss*100;
      line_color = CV_RGB(255,0,0);
      CvPoint pt1,pt2;
      pt1.x = x;
      pt1.y = y;
      pt2.x = static_cast<int>(x + scale_factor*cos(alpha_angle));
      pt2.y = static_cast<int>(y + scale_factor*sin(alpha_angle));

      cvLine( image, pt1, pt2 , line_color, 1, CV_AA, 0 );
      CvPoint p;
      p.x = (int) (pt2.x + 6 * cos(alpha_angle - pi / 4*3));
      p.y = (int) (pt2.y + 6 * sin(alpha_angle - pi / 4*3));
      cvLine( image, p, pt2, line_color, 1, CV_AA, 0 );
      p.x = (int) (pt2.x + 6 * cos(alpha_angle + pi / 4*3));
      p.y = (int) (pt2.y + 6 * sin(alpha_angle + pi / 4*3));
      cvLine( image, p, pt2, line_color, 1, CV_AA, 0 );

      /*
      line_color = CV_RGB(255,255,0);
      pt1.x = x-winsize;
      pt1.y = y-winsize;
      pt2.x = x+winsize;
      pt2.y = y+winsize;
      cvRectangle(image, pt1,pt2,line_color,1,CV_AA,0);
      */

     }
    }
   }
  }

 cvShowImage( "Contour", abs_img);
 cvShowImage( "Contour2", vely);

 cvReleaseImage(&velx);
 cvReleaseImage(&vely);
 cvWaitKey(20);
 
 return WW_OK;

}