Fixed bug in TLD::getVar

include/TLD.h

@@ -73,6 +73,8 @@ class TLD{
 //Integral Images
  cv::Mat iisum;
  cv::Mat iisqsum;
+ size_t step;
+ size_t esize;
  float var;
 
 //Training data
@@ -135,7 +137,7 @@ class TLD{
  void bbPoints(vector<cv::Point2f>& points, const BoundingBox& bb,int pts,int margin);
  void bbPredict(const vector<cv::Point2f>& points1,const vector<cv::Point2f>& points2,
  const BoundingBox& bb1,BoundingBox& bb2);
- float getVar(const BoundingBox& box,const cv::Mat& sum,const cv::Mat& sqsum);
+ double getVar(const BoundingBox& box,const cv::Mat& sum,const cv::Mat& sqsum);
  bool bbComp(const BoundingBox& bb1,const BoundingBox& bb2);
 };
 

src/TLD.cpp

@@ -47,8 +47,8 @@ void TLD::init(const Mat& frame1,const Rect& box){
  printf("Created %d bounding boxes\n",(int)grid.size());
  ///Preparation
  //allocation
- iisum.create(frame1.rows+1,frame1.cols+1,CV_64F);
- iisqsum.create(frame1.rows+1,frame1.cols+1,CV_64F);
+ iisum.create(frame1.rows+1,frame1.cols+1,CV_64FC1);
+ iisqsum.create(frame1.rows+1,frame1.cols+1,CV_64FC1);
  dconf.reserve(100);
  dbb.reserve(100);
  bbox_step =7;
@@ -169,6 +169,8 @@ void TLD::generateNegativeData(const Mat& frame){
  int idx;
  //Get Fern Features of the boxes with big variance (calculated using integral images)
  integral(frame,iisum,iisqsum);
+ esize = iisum.elemSize();
+ step = iisum.step;
  int a=0;
  int num = std::min((int)bad_boxes.size(),(int)bad_patches*100); //limits the size of bad_boxes to try
  printf("negative data generation started.\n");
@@ -193,23 +195,18 @@ void TLD::generateNegativeData(const Mat& frame){
  nEx.push_back(negExample);
  }
 }
-
-float TLD::getVar(const BoundingBox& box,const Mat& sum,const Mat& sqsum){
- /* NOTE: Optimize variance calculation?
- * for quick initialization of small matrices and/or super-fast element access
- * double m[3][3] = {{a, b, c}, {d, e, f}, {g, h, i}};
- * cv::Mat M = cv::Mat(3, 3, CV_64F, m).inv();
- */
- /* #define ELEM(type,start,step,size,xpos,ypos) *((type*)(start+step*(ypos)+(xpos)*size))
- * double v = ELEM(float,myMatrix.data,myMatrix.step(),myMatrix.elemSize,x,y); // Moderately speedy access
- * double v = ELEM(float,myMatrix.data,2560,4,x,y); // Even faster access
- */
- Point br(box.br().x+1,box.br().y+1);
- Point bl(box.x+1,box.y+box.height+1);
- Point tr(box.x+box.width+1,box.y+1);
- Point tl(box.x+1,box.y+1);
- float mean = (sum.at<double>(br)+sum.at<double>(tl)-sum.at<double>(tr)-sum.at<double>(bl))/box.area();
- float sqmean = (sqsum.at<double>(br)+sqsum.at<double>(tl)-sqsum.at<double>(tr)-sqsum.at<double>(bl))/box.area();
+#define ELEM(type,start,step,size,xpos,ypos) *((type*)(start+step*(ypos)+(xpos)*size))
+double TLD::getVar(const BoundingBox& box,const Mat& sum,const Mat& sqsum){
+ double brs = ELEM(double,sum.data,step,esize,box.x+box.width+1,box.y+box.height+1);
+ double bls = ELEM(double,sum.data,step,esize,box.x,box.y+box.height+1);
+ double trs = ELEM(double,sum.data,step,esize,box.x+box.width+1,box.y);
+ double tls = ELEM(double,sum.data,step,esize,box.x,box.y);
+ double brsq = ELEM(double,sqsum.data,step,esize,box.x+box.width+1,box.y+box.height+1);
+ double blsq = ELEM(double,sqsum.data,step,esize,box.x,box.y+box.height+1);
+ double trsq = ELEM(double,sqsum.data,step,esize,box.x+box.width+1,box.y);
+ double tlsq = ELEM(double,sqsum.data,step,esize,box.x,box.y);
+ float mean = (brs+tls-trs-bls)/box.area();
+ float sqmean = (brsq+tlsq-trsq-blsq)/box.area();
  return sqmean-mean*mean;
 }
 
@@ -316,8 +313,9 @@ void TLD::track(const Mat& img1, const Mat& img2,vector<Point2f>& points1,vector
  //Bounding box prediction
  bbPredict(points,points2,lastbox,tbb);
  //printf("predicted bbox %d %d %d %d\n",tbb.tl().x,tbb.tl().y,tbb.br().x,tbb.br().y);
- if (tracker.getFB()>10 ||tbb.tl().x < 0 || tbb.tl().y<0 || tbb.br().x > img2.cols || tbb.br().y > img2.rows){
+ if (tracker.getFB()>10 ||tbb.x < 0 || tbb.y<0 || tbb.br().x > img2.cols || tbb.br().y > img2.rows){
  tvalid =false; //too unstable prediction or bounding box out of image
+ tracked = false;
  return;
  }
  //Estimate Confidence and Validity
@@ -369,6 +367,8 @@ void TLD::detect(const cv::Mat& frame){
  double t = (double)getTickCount();
  Mat img(frame.rows,frame.cols,CV_8U);
  integral(frame,iisum,iisqsum);
+ esize = iisum.elemSize();
+ step = iisum.step;
  GaussianBlur(frame,img,Size(9,9),1.5);
  int numtrees = classifier.getNumStructs();
  float fern_th = classifier.getFernTh();