光流检测(optical flow)

 cvGoodFeaturesToTrack函数中count参数,虽然是输出参数,表示特征点的数目,但是这个初值设置至关重要。设置为0时,代码运行会错误。设置为1时,没有特征点输出,一般设置为最大特征点数目。(原因不明白)

 

cvGoodFeaturesToTrack,http://blog.csdn.net/moc062066/article/details/6634120,

cvFindCornerSubPix,http://blog.csdn.net/moc062066/article/details/6634961

两个函数的基础,接下来就是在视频中检测光流(optical flow),经常用的函数是cvCalcOpticalFlowPyrLK,函数说明如下;

Calculates the optical flow for a sparse feature set using the iterative Lucas-Kanade method with
pyramids.

//通过Lucas-Kanade方法与图像金字塔的结合,计算稀疏特征集合的光流

  1. <span style="font-size: 18px;">void cvCalcOpticalFlowPyrLK( 
  2.                             const CvArr* prev, 
  3.                             const CvArr* curr, 
  4.                             CvArr* prevPyr, 
  5.                             CvArr* currPyr, 
  6.                             const CvPoint2D32f* prevFeatures, 
  7.                             CvPoint2D32f* currFeatures, 
  8.                             int count, 
  9.                             CvSize winSize, 
  10.                             int level, 
  11.                             char* status, 
  12.                             float* track error, 
  13.                             CvTermCriteria criteria, 
  14.                             int flags );</span> 
void cvCalcOpticalFlowPyrLK(
							const CvArr* prev,
							const CvArr* curr,
							CvArr* prevPyr,
							CvArr* currPyr,
							const CvPoint2D32f* prevFeatures,
							CvPoint2D32f* currFeatures,
							int count,
							CvSize winSize,
							int level,
							char* status,
							float* track error,
							CvTermCriteria criteria,
							int flags );

prev First frame, at time t  //取t时刻为第一帧

curr Second frame, at time t + dt //第二帧出现在 t + dt时刻

prevPyr Buffer for the pyramid for the first frame. If the pointer is not NULL , the buffer must
have a sufficient size to store the pyramid from level 1 to level level ; the total size of
(image width+8)*image height/3 bytes is sufficient

//第一帧的图像金字塔的缓存之处。如果该指针不为空,该buffer必须有足够的空间来存储从第1层到第level 层的图像金字塔;prevPyr指针所指的图像/矩阵的大小为(image width+8)* (image height/3) 就足够了。

currPyr Similar to prevPyr, used for the second frame //同上

prevFeatures Array of points for which the flow needs to be found //在数组中定义(当前帧中的)那些点是要在(下一帧)检测的

currFeatures Array of 2D points containing the calculated new positions of the input features
in the second image

//一个二维的点数组,用于存放输入的特征(就是prevFeatures)在第二帧中的新位置

count Number of feature points//特征点的数目

winSize Size of the search window of each pyramid level //每一层金字塔所有的搜索窗口的大小

level Maximal pyramid level number. If 0 , pyramids are not used (single level), if 1 , two levels
are used, etc

//最多有多少层金字塔。如果是0,就不用图像金字塔,如果是1,就有两层,以此类推。

status Array. Every element of the array is set to 1 if the flow for the corresponding feature has
been found, 0 otherwise

//是一个数组,对应点在第二帧中找到,那该位置就值为1,找不到就值为0.

track error Array of double numbers containing the difference between patches around the
original and moved points. Optional parameter; can be NULL
criteria Specifies when the iteration process of finding the flow for each point on each pyramid
level should be stopped

flags Miscellaneous flags:

CV LKFLOWPyr A READY pyramid for the first frame is precalculated before the call
CV LKFLOWPyr B READY pyramid for the second frame is precalculated before the call
CV LKFLOW INITIAL GUESSES array B contains initial coordinates of features before the
function call

cvCalcOpticalFlowPyrLK函数中最后一个标志位,FLAG在视频处理中很重要。要提高速度,不重复计算金字塔,这里需要设置FLAG.即保存前一帧的金字塔数据和特征点,做为下一帧的初始帧。

 

  1. //cvCaclOpticalFlowPyrLk_demo 
  2. //mochen 
  3. //2011年7月26日20:23:42 
  4.  
  5. #include <stdio.h> 
  6. #include "cv.h" 
  7. #include "cxcore.h" 
  8. #include "highgui.h" 
  9.  
  10. #pragma comment(lib, "opencv_core220d.lib") 
  11. #pragma comment(lib, "opencv_highgui220d.lib") 
  12. #pragma comment(lib, "opencv_imgproc220d.lib") 
  13. #pragma comment(lib, "opencv_calib3d220d.lib") 
  14. #pragma comment(lib, "opencv_features2d220d.lib") 
  15. #pragma comment(lib, "opencv_contrib220d.lib") 
  16. #pragma comment(lib, "opencv_ml220d.lib")  
  17. #pragma comment(lib, "opencv_video220d.lib") 
  18.  
  19.  
  20. #if 0 
  21. void cvCalcOpticalFlowPyrLK( 
  22.                             const CvArr* prev, 
  23.                             const CvArr* curr, 
  24.                             CvArr* prevPyr, 
  25.                             CvArr* currPyr, 
  26.                             const CvPoint2D32f* prevFeatures, 
  27.                             CvPoint2D32f* currFeatures, 
  28.                             int count, 
  29.                             CvSize winSize, 
  30.                             int level, 
  31.                             char* status, 
  32.                             float* track error, 
  33.                             CvTermCriteria criteria, 
  34.                             int flags ); 
  35. #endif 
  36.  
  37. const int MAX_CORNERS = 1000 ; 
  38.  
  39. int main(int argc,char** argv) 
  40.     while ( 1 ) 
  41.     { 
  42.         //use webcam  
  43.         CvCapture* cam = cvCaptureFromCAM( CV_CAP_ANY ) ; 
  44.         assert( NULL != cam ) ; 
  45.  
  46.         //get a color image  
  47.         IplImage* frame = cvQueryFrame(cam) ; 
  48.  
  49.         CvSize img_sz = cvGetSize(frame); 
  50.         const int win_size = 10 ; 
  51.  
  52.         //convert the image to grey image 
  53.         IplImage* frame_prev = cvQueryFrame(cam) ; 
  54.         IplImage* img_prev = cvCreateImage(img_sz,IPL_DEPTH_8U,1) ; 
  55.         cvCvtColor( frame_prev,img_prev ,CV_BGR2GRAY); 
  56.  
  57.         //convert the image to grey image 
  58.         IplImage* frame_cur = cvQueryFrame(cam) ; 
  59.         IplImage* img_curr = cvCreateImage(img_sz,IPL_DEPTH_8U,1) ; 
  60.         cvCvtColor( frame_cur,img_curr ,CV_BGR2GRAY); 
  61.  
  62.         //create a imge to display result 
  63.         IplImage* img_res = cvCreateImage(img_sz,IPL_DEPTH_8U,1) ; 
  64.         for ( int y = 0 ; y < img_sz.height ; ++y ) 
  65.         { 
  66.             uchar* ptr = (uchar*)( img_res->imageData + y * img_res->widthStep ) ; 
  67.             for ( int x = 0 ; x <img_res->width; ++x ) 
  68.             { 
  69.                 ptr[x] = 255 ; 
  70.             } 
  71.         } 
  72.  
  73.         //get good features  
  74.         IplImage* img_eig = cvCreateImage(img_sz,IPL_DEPTH_32F,1) ; 
  75.         IplImage* img_temp = cvCreateImage(img_sz,IPL_DEPTH_32F,1) ; 
  76.         int corner_count = MAX_CORNERS ; 
  77.         CvPoint2D32f*  features_prev = new CvPoint2D32f[MAX_CORNERS] ; 
  78.  
  79.         cvGoodFeaturesToTrack( 
  80.             img_prev, 
  81.             img_eig, 
  82.             img_temp, 
  83.             features_prev, 
  84.             &corner_count, 
  85.             0.01, 
  86.             5.0, 
  87.             0, 
  88.             3, 
  89.             0, 
  90.             0.4 
  91.             ); 
  92.  
  93.         cvFindCornerSubPix( 
  94.             img_prev, 
  95.             features_prev, 
  96.             corner_count, 
  97.             cvSize(win_size,win_size), 
  98.             cvSize(-1,-1), 
  99.             cvTermCriteria(CV_TERMCRIT_EPS | CV_TERMCRIT_ITER,20,0.03) 
  100.             ); 
  101.  
  102.         // L-K  
  103.         char feature_found[ MAX_CORNERS ] ; 
  104.         float feature_errors[ MAX_CORNERS ] ; 
  105.  
  106.         CvSize pyr_sz = cvSize( frame->width + 8 ,frame->height / 3 ) ; 
  107.  
  108.         IplImage* pyr_prev = cvCreateImage(img_sz,IPL_DEPTH_32F,1) ; 
  109.         IplImage* pyr_cur = cvCreateImage(img_sz,IPL_DEPTH_32F,1) ; 
  110.         CvPoint2D32f*  features_cur = new CvPoint2D32f[ MAX_CORNERS ] ; 
  111.  
  112.         cvCalcOpticalFlowPyrLK( 
  113.             img_prev, 
  114.             img_curr, 
  115.             pyr_prev, 
  116.             pyr_cur, 
  117.             features_prev, 
  118.             features_cur, 
  119.             corner_count, 
  120.             cvSize(win_size,win_size), 
  121.             5, 
  122.             feature_found, 
  123.             feature_errors, 
  124.             cvTermCriteria(CV_TERMCRIT_EPS | CV_TERMCRIT_ITER,20,0.3), 
  125.             0 
  126.             ); 
  127.  
  128.         for ( int i = 0 ; i < corner_count ; i++) 
  129.         { 
  130.             if ( 0 == feature_found[i] || feature_errors[i] > 550 ) 
  131.             { 
  132.                 printf("error is %f \n" , feature_errors[i] ) ; 
  133.                 continue
  134.             } 
  135.  
  136.             printf("find it !\n") ; 
  137.  
  138.             CvPoint pt_prev = cvPoint( features_prev[i].x , features_prev[i].y ) ; 
  139.             CvPoint pt_cur = cvPoint( features_cur[i].x , features_cur[i].y ) ; 
  140.  
  141.             cvLine( img_res,pt_prev,pt_cur,CV_RGB( 255,0,0),2 ); 
  142.         } 
  143.  
  144.         const char* window_prev ="img_prev"
  145.         const char* window_curr ="img_curr"
  146.         const char* window_res ="result"
  147.         cvNamedWindow(window_prev,CV_WINDOW_AUTOSIZE); 
  148.         cvNamedWindow(window_curr,CV_WINDOW_AUTOSIZE); 
  149.         cvNamedWindow(window_res,CV_WINDOW_AUTOSIZE); 
  150.  
  151.         cvShowImage( window_prev,img_prev ); 
  152.         cvShowImage( window_curr,img_curr ); 
  153.         cvShowImage( window_res,img_res ); 
  154.  
  155.         char opt = cvWaitKey( 10000 ) ; 
  156.         if ( 27 == opt ) 
  157.         { 
  158.             break
  159.         } 
  160.  
  161.         cvReleaseCapture( &cam ); 
  162.         cvReleaseImage( &img_curr ); 
  163.         cvReleaseImage( &img_eig ); 
  164.         cvReleaseImage( &img_prev ); 
  165.         cvReleaseImage( &img_res ); 
  166.         cvReleaseImage( &img_temp ); 
  167.         cvDestroyAllWindows() ; 
  168.     } 
  169.  
  170.  
  171.     return 0 ; 

 

下面一段代码为视频处理,设置FLAG

int CMotionOptic::OpticalFlow()

  CvPoint2D32f *m_pPreConners1 = new CvPoint2D32f[MAX_CONNERS];
  CvPoint2D32f *m_pCurConners1 = new CvPoint2D32f[MAX_CONNERS];

 char aFeatureFound[MAX_CONNERS];
 float aTrackErrors[MAX_CONNERS];

  IplImage* m_pPyrCurImg1 = cvCreateImage(cvSize(m_iWidth,m_iHeight), IPL_DEPTH_32F,1); 
  IplImage *m_pPyrPreImg1 = cvCreateImage(cvSize(m_iWidth,m_iHeight), IPL_DEPTH_32F,1);

 //int m_iConnersCnt1 = MAX_CONNERS;

 if(1 == m_iStartFrameCnt)
 {
  
  cvGoodFeaturesToTrack(m_pPreImg,m_pEigImg,m_pTmpImg,m_pPreConners,&m_iConnersCnt,0.01,5.0,0,3,0,0.04);

   cvFindCornerSubPix(m_pPreImg,m_pPreConners,m_iConnersCnt,cvSize(10,10),cvSize(-1,-1),
    cvTermCriteria(CV_TERMCRIT_EPS|CV_TERMCRIT_ITER, 200,0.03));

   cvCalcOpticalFlowPyrLK(m_pPreImg, m_pCurImg, m_pPyrPreImg, m_pPyrCurImg, m_pPreConners,
    m_pCurConners,m_iConnersCnt,cvSize(10,10),5,aFeatureFound,aTrackErrors,
    cvTermCriteria(CV_TERMCRIT_EPS|CV_TERMCRIT_ITER, 200, 0.03),0); 
 }
 else
 {
  cvCalcOpticalFlowPyrLK(m_pPreImg, m_pCurImg, m_pPyrPreImg, m_pPyrCurImg, m_pPreConners,
   m_pCurConners,m_iConnersCnt,cvSize(10,10),5,aFeatureFound,aTrackErrors,
   cvTermCriteria(CV_TERMCRIT_EPS|CV_TERMCRIT_ITER, 200, 0.03),CV_LKFLOW_PYR_A_READY);  
 } 

 
 for(int iLoop = 0;iLoop < m_iConnersCnt; iLoop++)
 {
  if((0 == aFeatureFound[iLoop])||(aTrackErrors[iLoop] > 550))
  {
   printf("Error is %f",aTrackErrors[iLoop]);
   continue;
  }

  CvPoint pt0 = cvPoint(cvRound(m_pPreConners[iLoop].x), cvRound(m_pPreConners[iLoop].y));
  CvPoint pt1 = cvPoint(cvRound(m_pCurConners[iLoop].x), cvRound(m_pCurConners[iLoop].y));

  if((abs(pt0.x - pt1.x) >= 1)||(abs(pt0.y - pt1.y) >= 1))
  {
   cvLine(m_pCurImg,pt0,pt1,cvScalar(255,0,0),2);
   cvLine(m_pPreImg,pt0,pt1,cvScalar(255,0,0),2);
  }
 }

 cvShowImage("PreOptica", m_pPreImg);
 cvShowImage("CurOptica", m_pCurImg);

 memcpy(m_pPyrPreImg->imageData, m_pPyrCurImg->imageData, (m_iWidth+8)*(m_iHeight/3)*4);
 memcpy(m_pPreConners, m_pCurConners, m_iConnersCnt*sizeof(CvPoint2D32f)); 

 return 0;
}

 

 

 

没有更多推荐了,返回首页