光流法-运动目标的检测二（opencv学习）

 

光流法-运动目标的检测二（opencv学习）

2011-11-04 11:05

正如上一篇文章所述， cvCalcOpticalFlowPyrLK 需要确定特征点。          本程序，通过帧差获得运动的点作为特征点。          本程序原本的目的是计算运动点的速度，通过修正可以进行运动跟踪。   #include <cv.h> #include <highgui.h> #include <iostream> using namespace std; int const MAX_CORNERS = 1000; int main (int argc, char **argv) {   CvCapture* capture = 0;  capture = cvCaptureFromCAM(  CV_CAP_ANY );  //get frame  IplImage *src_img1;  //the previous frame (gray)  IplImage *src_img2;  //the current frame(gray)  IplImage *dst_img;   //the result  IplImage *cur_img;    IplImage *pre_img;  CvPoint2D32f * move_old_point = new CvPoint2D32f[ MAX_CORNERS];  CvPoint2D32f * move_new_point = new CvPoint2D32f[ MAX_CORNERS];  char *features_found = new char[MAX_CORNERS];  float *features_error = new float[MAX_CORNERS];  CvTermCriteria criteria;  criteria = cvTermCriteria (CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, 64, 0.01);  while(1)  {   int i,j;   int dx, dy;   int p = 0;   int rows, cols;   int countn = MAX_CORNERS;   pre_img = cvQueryFrame(capture);   CvSize img_sz = cvGetSize(pre_img);   src_img1 = cvCreateImage(img_sz, IPL_DEPTH_8U, 1);   cvCvtColor(pre_img, src_img1, CV_RGB2GRAY);   cur_img = cvQueryFrame(capture);   src_img2 = cvCreateImage(img_sz, IPL_DEPTH_8U, 1);   cvCvtColor(cur_img, src_img2, CV_RGB2GRAY);   dst_img = (IplImage *)cvClone(cur_img);   IplImage *move_img = cvCreateImage(img_sz, IPL_DEPTH_8U, 1);   cvZero(move_img);   //cvAbsDiff(src_img1, src_img2,move_img);   cols = src_img1->width;    rows = src_img1->height;   for (i = 0; i <cols; i++)   {    for (j = 0; j<rows; j++)    {     double a = abs(cvGet2D(src_img1, j, i).val[0]-cvGet2D(src_img2, j, i).val[0]);     CvScalar b = cvScalar(a, 0, 0,0);     cvSet2D(move_img, j, i,b);     if (a>40)     {      if (p<MAX_CORNERS-1)      {       int d = ++p;       move_old_point[d].x = i;       move_old_point[d].y = j;      }     }    }   }         cvNamedWindow("moving object", 1);   cvShowImage("moving object", move_img);   CvSize Pyrsize = cvSize(src_img1->width +8, src_img1->height/3);   IplImage * pyrA = cvCreateImage(Pyrsize, IPL_DEPTH_32F, 1); //pyrA是需要寻找的点，不是没有初始化的   IplImage * pyrB = cvCreateImage(Pyrsize, IPL_DEPTH_32F, 1);   cvCalcOpticalFlowPyrLK(src_img1,         src_img2,         pyrA,         pyrB,         move_old_point,         move_new_point,         countn,         cvSize(10, 10),         3,         features_found,         features_error,         criteria,         0         );   for (i = 0; i < countn; i++)   {      int x1 = (int)move_new_point[i].x;    int x2 = (int)move_old_point[i].x;    int y1 = (int)move_new_point[i].y;    int y2 = (int)move_old_point[i].y;      dx =(int) abs(x1 - x2) ;    dy = (int)abs(y1 - y2);    if (dx >= 5&& dy >= 5)    {     cvLine (dst_img, cvPoint(x2, y2),cvPoint(x2+5, y2+5) , CV_RGB (255, 0, 0), 1, CV_AA, 0);    }   }   cvNamedWindow ("ImagePyrLK", 1);    cvShowImage ("ImagePyrLK", dst_img);   cvWaitKey (1);   cvReleaseImage (&dst_img);   cvReleaseImage(&pyrA);   cvReleaseImage(&pyrB);   cvReleaseImage(&move_img);  }  cvDestroyWindow("moving object");  cvDestroyWindow ("ImagePyrLK");   cvReleaseImage (&src_img1);  cvReleaseImage (&src_img2);  cvReleaseImage (&pre_img);  cvReleaseImage (&cur_img);  return 0; }