Vadim

modules/tracking/src/TLD.cpp

@@ -181,23 +181,23 @@ double variance(const Mat& img){
     p2/=(img.cols*img.rows);
     return p2-p*p;
 }
-double variance(Mat_<unsigned int>& intImgP,Mat_<unsigned int>& intImgP2,Rect box){
+double variance(Mat_<double>& intImgP,Mat_<double>& intImgP2,Rect box){
     int x=(box.x),y=(box.y),width=(box.width),height=(box.height);
-    CV_Assert(0<=x && (x+width)<=intImgP.cols && (x+width)<=intImgP2.cols);
-    CV_Assert(0<=y && (y+height)<=intImgP.rows && (y+height)<=intImgP2.rows);
+    CV_Assert(0<=x && (x+width)<intImgP.cols && (x+width)<intImgP2.cols);
+    CV_Assert(0<=y && (y+height)<intImgP.rows && (y+height)<intImgP2.rows);
     double p=0,p2=0;
-    unsigned int A,B,C,D;
+    double A,B,C,D;
 
-    A=((y>0&&x>0)?intImgP(y-1,x-1):0);
-    B=((y>0)?intImgP(y-1,x+width-1):0);
-    C=((x>0)?intImgP(y+height-1,x-1):0);
-    D=intImgP(y+height-1,x+width-1);
+    A=intImgP(y,x);
+    B=intImgP(y,x+width);
+    C=intImgP(y+height,x);
+    D=intImgP(y+height,x+width);
     p=(0.0+A+D-B-C)/(width*height);
 
-    A=((y>0&&x>0)?intImgP2(y-1,x-1):0);
-    B=((y>0)?intImgP2(y-1,x+width-1):0);
-    C=((x>0)?intImgP2(y+height-1,x-1):0);
-    D=intImgP2(y+height-1,x+width-1);
+    A=intImgP2(y,x);
+    B=intImgP2(y,x+width);
+    C=intImgP2(y+height,x);
+    D=intImgP2(y+height,x+width);
     p2=(0.0+(D-B)-(C-A))/(width*height);
 
     return p2-p*p;
@@ -253,11 +253,10 @@ void resample(const Mat& img,const Rect2d& r2,Mat_<uchar>& samples){
     M(0,0)=(float)(samples.cols/r2.width); M(0,1)=0.0f; M(0,2)=(float)(-r2.x*samples.cols/r2.width);
     M(1,0)=0.0f; M(1,1)=(float)(samples.rows/r2.height); M(1,2)=(float)(-r2.y*samples.rows/r2.height);
     warpAffine(img,samples,M,samples.size());
-
 }
 
 //other stuff
-void TLDEnsembleClassifier::stepPrefSuff(uchar* arr,int len){
+void TLDEnsembleClassifier::stepPrefSuff(std::vector<uchar>& arr,int len){
     int gridSize=getGridSize();
 #if 0
         int step=len/(gridSize-1), pref=(len-step*(gridSize-1))/2;
@@ -270,7 +269,7 @@ void TLDEnsembleClassifier::stepPrefSuff(uchar* arr,int len){
         int smallStep=quo,bigStep=quo+1;
         int bigOnes=rem,smallOnes=gridSize-bigOnes-1;
         int bigOnes_front=bigOnes/2,bigOnes_back=bigOnes-bigOnes_front;
-        for(int i=0;i<(int)(sizeof(x1)/sizeof(x1[0]));i++){
+        for(int i=0;i<(int)arr.size();i++){
             if(arr[i]<bigOnes_back){
                 arr[i]=(uchar)(arr[i]*bigStep+arr[i]);
                 continue;
@@ -287,12 +286,23 @@ void TLDEnsembleClassifier::stepPrefSuff(uchar* arr,int len){
         }
 #endif
 }
-TLDEnsembleClassifier::TLDEnsembleClassifier(int ordinal,Size size){
+TLDEnsembleClassifier::TLDEnsembleClassifier(int ordinal,Size size,int measurePerClassifier){
+    x1=std::vector<uchar>(measurePerClassifier,0);
+    x2=std::vector<uchar>(measurePerClassifier,0);
+    y1=std::vector<uchar>(measurePerClassifier,0);
+    y2=std::vector<uchar>(measurePerClassifier,0);
+
     preinit(ordinal);
+
     stepPrefSuff(x1,size.width);
     stepPrefSuff(x2,size.width);
     stepPrefSuff(y1,size.height);
     stepPrefSuff(y2,size.height);
+
+    int posSize=1;
+    for(int i=0;i<measurePerClassifier;i++)posSize*=2;
+    pos=std::vector<unsigned int>(posSize,0);
+    neg=std::vector<unsigned int>(posSize,0);
 }
 void TLDEnsembleClassifier::integrate(Mat_<uchar> patch,bool isPositive){
     unsigned short int position=code(patch.data,(int)patch.step[0]);
@@ -313,18 +323,13 @@ double TLDEnsembleClassifier::posteriorProbability(const uchar* data,int rowstep
 }
 unsigned short int TLDEnsembleClassifier::code(const uchar* data,int rowstep)const{
     unsigned short int position=0;
-    //char codeS[20];
-    for(int i=0;i<(int)(sizeof(x1)/sizeof(x1[0]));i++){
+    for(int i=0;i<(int)x1.size();i++){
         position=position<<1;
         if(*(data+rowstep*y1[i]+x1[i])<*(data+rowstep*y2[i]+x2[i])){
             position++;
-            //codeS[i]='o';
         }else{
-            //codeS[i]='x';
         }
     }
-    //codeS[13]='\0';
-    //dprintf(("integrate with code %s\n",codeS));
     return position;
 }
 

modules/tracking/src/TLD.hpp

@@ -79,7 +79,7 @@ double overlap(const Rect2d& r1,const Rect2d& r2);
 void resample(const Mat& img,const RotatedRect& r2,Mat_<uchar>& samples);
 void resample(const Mat& img,const Rect2d& r2,Mat_<uchar>& samples);
 double variance(const Mat& img);
-double variance(Mat_<unsigned int>& intImgP,Mat_<unsigned int>& intImgP2,Rect box);
+double variance(Mat_<double>& intImgP,Mat_<double>& intImgP2,Rect box);
 double NCC(Mat_<uchar> patch1,Mat_<uchar> patch2);
 void getClosestN(std::vector<Rect2d>& scanGrid,Rect2d bBox,int n,std::vector<Rect2d>& res);
 double scaleAndBlur(const Mat& originalImg,int scale,Mat& scaledImg,Mat& blurredImg,Size GaussBlurKernelSize);
@@ -87,17 +87,17 @@ unsigned int getMedian(const std::vector<unsigned int>& values, int size=-1);
 
 class TLDEnsembleClassifier{
 public:
-    TLDEnsembleClassifier(int ordinal,Size size);
+    TLDEnsembleClassifier(int ordinal,Size size,int measurePerClassifier);
     void integrate(Mat_<uchar> patch,bool isPositive);
     double posteriorProbability(const uchar* data,int rowstep)const;
     static int getMaxOrdinal();
 private:
     static int getGridSize();
-    inline void stepPrefSuff(uchar* arr,int len);
+    inline void stepPrefSuff(std::vector<uchar>& arr,int len);
     void preinit(int ordinal);
     unsigned short int code(const uchar* data,int rowstep)const;
-    unsigned int pos[8192],neg[8192];//8192=2^13
-    uchar x1[13],y1[13],x2[13],y2[13];
+    std::vector<unsigned int> pos,neg;
+    std::vector<uchar> x1,y1,x2,y2;
 };
 
 class TrackerProxy : public TrackerTLD::Private{

modules/tracking/src/trackerTLD.cpp

@@ -52,6 +52,7 @@
 #define CORE_THRESHOLD 0.5
 #define NEG_EXAMPLES_IN_INIT_MODEL 300
 #define MAX_EXAMPLES_IN_MODEL 500
+#define MEASURES_PER_CLASSIFIER 13
 static const cv::Size GaussBlurKernelSize(3,3);
 
 using namespace cv;
@@ -65,19 +66,14 @@ using namespace tld;
  *      THETA_NN 0.5<->0.6 dramatic change vs video 6 !!
  * TODO:
  *      schoolPC: codec, libopencv-dev
- *      fix pushbot ->pick commits -> compare_branches->all in 1
- *      ||video(0.5<->0.6) --> debug if box size is less than 20 --> (remove ensemble self-loop) --> (try inter_area)
+ *      fix pushbot ->pick commits -> compare_branches->all in 1->resubmit
+ *      ||video(0.5<->0.6) --> debug if box size is less than 20 --> (remove ensemble self-loop) --> (try inter_area for resize)
  *      perfect PN
  *
  *      vadim:
  *
- *      private members
- *
- *      cv::integral
- *
- *      13 as enum
- *
  *      blurred in TrackerTLDModel()
+ *      private members
  *
  *      warpAffine -- ?
 */
@@ -129,8 +125,8 @@ public:
 protected:
     friend class MyMouseCallbackDEBUG;
     Ptr<TrackerModel> model;
-    void computeIntegralImages(const Mat& img,Mat_<unsigned int>& intImgP,Mat_<unsigned int>& intImgP2);
-    bool patchVariance(Mat_<unsigned int>& intImgP,Mat_<unsigned int>& intImgP2,double originalVariance,Point pt,Size size);
+    inline void computeIntegralImages(const Mat& img,Mat_<double>& intImgP,Mat_<double>& intImgP2){integral(img,intImgP,intImgP2,CV_64F);}
+    inline bool patchVariance(Mat_<double>& intImgP,Mat_<double>& intImgP2,double originalVariance,Point pt,Size size);
     bool ensembleClassifier(const uchar* data,int rowstep){return ensembleClassifierNum(data,rowstep)>0.5;}
     double ensembleClassifierNum(const uchar* data,int rowstep);
     TrackerTLD::Params params_;
@@ -406,7 +402,7 @@ timeStampPositiveNext(0),timeStampNegativeNext(0),params_(params){
 
     Mat_<uchar> blurredPatch(minSize);
     for(int i=0,howMany=TLDEnsembleClassifier::getMaxOrdinal();i<howMany;i++){
-        classifiers.push_back(TLDEnsembleClassifier(i+1,minSize));
+        classifiers.push_back(TLDEnsembleClassifier(i+1,minSize,MEASURES_PER_CLASSIFIER));
     }
 
     positiveExamples.reserve(200);
@@ -507,7 +503,7 @@ bool TLDDetector::detect(const Mat& img,const Mat& imgBlurred,Rect2d& res,std::v
     Rect2d maxScRect;
     START_TICK("detector");
     do{
-        Mat_<unsigned int> intImgP(resized_img.rows,resized_img.cols),intImgP2(resized_img.rows,resized_img.cols);
+        Mat_<double> intImgP,intImgP2;
         computeIntegralImages(resized_img,intImgP,intImgP2);
 
         for(int i=0,imax=cvFloor((0.0+resized_img.cols-initSize.width)/dx);i<imax;i++){
@@ -595,22 +591,7 @@ bool TLDDetector::detect(const Mat& img,const Mat& imgBlurred,Rect2d& res,std::v
     return true;
 }
 
-void TLDDetector::computeIntegralImages(const Mat& img,Mat_<unsigned int>& intImgP,Mat_<unsigned int>& intImgP2){
-    intImgP(0,0)=img.at<uchar>(0,0);
-    for(int j=1;j<intImgP.cols;j++){intImgP(0,j)=intImgP(0,j-1)+img.at<uchar>(0,j);}
-    for(int i=1;i<intImgP.rows;i++){intImgP(i,0)=intImgP(i-1,0)+img.at<uchar>(i,0);}
-    for(int i=1;i<intImgP.rows;i++){for(int j=1;j<intImgP.cols;j++){
-            intImgP(i,j)=intImgP(i,j-1)-intImgP(i-1,j-1)+intImgP(i-1,j)+img.at<uchar>(i,j);}}
-
-    unsigned int p;
-    p=img.at<uchar>(0,0);intImgP2(0,0)=p*p;
-    for(int j=1;j<intImgP2.cols;j++){p=img.at<uchar>(0,j);intImgP2(0,j)=intImgP2(0,j-1)+p*p;}
-    for(int i=1;i<intImgP2.rows;i++){p=img.at<uchar>(i,0);intImgP2(i,0)=intImgP2(i-1,0)+p*p;}
-    for(int i=1;i<intImgP2.rows;i++){for(int j=1;j<intImgP2.cols;j++){p=img.at<uchar>(i,j);
-            intImgP2(i,j)=intImgP2(i,j-1)-intImgP2(i-1,j-1)+intImgP2(i-1,j)+p*p;}}
-}
-
-bool TLDDetector::patchVariance(Mat_<unsigned int>& intImgP,Mat_<unsigned int>& intImgP2,double originalVariance,Point pt,Size size){
+bool TLDDetector::patchVariance(Mat_<double>& intImgP,Mat_<double>& intImgP2,double originalVariance,Point pt,Size size){
     return variance(intImgP,intImgP2,Rect(pt.x,pt.y,size.width,size.height)) >= 0.5*originalVariance;
 }
 
@@ -818,7 +799,7 @@ void MyMouseCallbackDEBUG::onMouse( int event, int x, int y){
         resize(img_,resized_img,size);
         resize(imgBlurred_,blurred_img,size);
 
-        Mat_<unsigned int> intImgP(resized_img.rows,resized_img.cols),intImgP2(resized_img.rows,resized_img.cols);
+        Mat_<double> intImgP,intImgP2;
         detector_->computeIntegralImages(resized_img,intImgP,intImgP2);
 
         int dx=initSize.width/10, dy=initSize.height/10,