Merge pull request #55 from nailbiter/luca

TLD

modules/tracking/doc/tracker_algorithms.rst

@@ -9,6 +9,10 @@ The following algorithms are implemented at the moment.
 
 .. [OLB] H Grabner, M Grabner, and H Bischof, Real-time tracking via on-line boosting, In Proc. BMVC, volume 1, pages 47– 56, 2006
 
+.. [MedianFlow] Z. Kalal, K. Mikolajczyk, and J. Matas, “Forward-Backward Error: Automatic Detection of Tracking Failures,” International Conference on Pattern Recognition, 2010, pp. 23-26. 
+
+.. [TLD] Z. Kalal, K. Mikolajczyk, and J. Matas, “Tracking-Learning-Detection,” Pattern Analysis and Machine Intelligence 2011.
+
 TrackerBoosting
 ---------------
 
@@ -63,7 +67,7 @@ Constructor
     :param parameters: BOOSTING parameters :ocv:struct:`TrackerBoosting::Params`
 
 TrackerMIL
-----------
+----------------------
 
 The MIL algorithm trains a classifier in an online manner to separate the object from the background. Multiple Instance Learning avoids the drift problem for a robust tracking. The implementation is based on [MIL]_.
 
@@ -118,3 +122,105 @@ Constructor
 .. ocv:function:: Ptr<trackerMIL> TrackerMIL::createTracker(const trackerMIL::Params &parameters=trackerMIL::Params())
 
     :param parameters: MIL parameters :ocv:struct:`TrackerMIL::Params`
+
+TrackerMedianFlow
+----------------------
+
+Implementation of a paper "Forward-Backward Error: Automatic Detection of Tracking Failures" by Z. Kalal, K. Mikolajczyk 
+and Jiri Matas. The implementation is based on [MedianFlow]_.
+
+The tracker is suitable for very smooth and predictable movements when object is visible throughout the whole sequence. It's quite and
+accurate for this type of problems (in particular, it was shown by authors to outperform MIL). During the implementation period the code at
+http://www.aonsquared.co.uk/node/5, the courtesy of the author Arthur Amarra, was used for the reference purpose.
+
+.. ocv:class:: TrackerMedianFlow
+
+Implementation of TrackerMedianFlow from :ocv:class:`Tracker`::
+
+    class CV_EXPORTS_W TrackerMedianFlow : public Tracker
+    {
+     public:
+      void read( const FileNode& fn );
+      void write( FileStorage& fs ) const;
+      static Ptr<trackerMedianFlow> createTracker(const trackerMedianFlow::Params &parameters=trackerMedianFlow::Params());
+      virtual ~trackerMedianFlow(){};
+
+     protected:
+      bool initImpl( const Mat& image, const Rect2d& boundingBox );
+      bool updateImpl( const Mat& image, Rect2d& boundingBox );
+    };
+
+TrackerMedianFlow::Params
+------------------------------------
+
+.. ocv:struct:: TrackerMedianFlow::Params
+
+List of MedianFlow parameters::
+
+   struct CV_EXPORTS Params
+   {
+    Params();
+    int pointsInGrid; //square root of number of keypoints used; increase it to trade
+                      //accurateness for speed; default value is sensible and recommended
+                      
+    void read( const FileNode& fn );
+    void write( FileStorage& fs ) const;
+   };
+
+TrackerMedianFlow::createTracker
+-----------------------------------
+
+Constructor
+
+.. ocv:function:: Ptr<trackerMedianFlow> TrackerMedianFlow::createTracker(const trackerMedianFlow::Params &parameters=trackerMedianFlow::Params())
+
+    :param parameters: Median Flow parameters :ocv:struct:`TrackerMedianFlow::Params`
+
+TrackerTLD
+----------------------
+
+TLD is a novel tracking framework that explicitly decomposes the long-term tracking task into tracking, learning and detection. The tracker follows the object from frame to frame. The detector localizes all appearances that have been observed so far and corrects the tracker if necessary. The learning estimates detector’s errors and updates it to avoid these errors in the future. The implementation is based on [TLD]_.
+
+The Median Flow algorithm (see above) was chosen as a tracking component in this implementation, following authors. Tracker is supposed to be able
+to handle rapid motions, partial occlusions, object absence etc.
+
+.. ocv:class:: TrackerTLD
+
+Implementation of TrackerTLD from :ocv:class:`Tracker`::
+
+    class CV_EXPORTS_W TrackerTLD : public Tracker
+    {
+     public:
+      void read( const FileNode& fn );
+      void write( FileStorage& fs ) const;
+      static Ptr<trackerTLD> createTracker(const trackerTLD::Params &parameters=trackerTLD::Params());
+      virtual ~trackerTLD(){};
+
+     protected:
+      bool initImpl( const Mat& image, const Rect2d& boundingBox );
+      bool updateImpl( const Mat& image, Rect2d& boundingBox );
+    };
+
+TrackerTLD::Params
+------------------------
+
+.. ocv:struct:: TrackerTLD::Params
+
+List of TLD parameters::
+
+   struct CV_EXPORTS Params
+   {
+    Params();
+
+    void read( const FileNode& fn );
+    void write( FileStorage& fs ) const;
+   };
+
+TrackerTLD::createTracker
+-------------------------------
+
+Constructor
+
+.. ocv:function:: Ptr<trackerTLD> TrackerTLD::createTracker(const trackerTLD::Params &parameters=trackerTLD::Params())
+
+    :param parameters: TLD parameters :ocv:struct:`TrackerTLD::Params`

modules/tracking/include/opencv2/tracking/tracker.hpp

@@ -1017,7 +1017,8 @@ class CV_EXPORTS_W TrackerMedianFlow : public Tracker
   struct CV_EXPORTS Params
   {
     Params();
-    int pointsInGrid;
+    int pointsInGrid; //square root of number of keypoints used; increase it to trade
+                      //accurateness for speed; default value is sensible and recommended
     void read( const FileNode& /*fn*/ );
     void write( FileStorage& /*fs*/ ) const;
   };

modules/tracking/samples/benchmark.cpp

@@ -7,9 +7,9 @@
 #include <cstring>
 #include <climits>
 
-#define CMDLINEMAX 10
-#define ASSESS_TILL 100
-#define LINEMAX 40
+const int CMDLINEMAX = 30;
+const int ASSESS_TILL = 100;
+const int LINEMAX = 40;
 
 using namespace std;
 using namespace cv;
@@ -20,7 +20,8 @@ using namespace cv;
 
 static Mat image;
 static bool paused;
-vector<Scalar> palette;
+static bool saveImageKey;
+static vector<Scalar> palette;
 
 void print_table(char* videos[],int videoNum,char* algorithms[],int algNum,const vector<vector<char*> >& results,char* tableName);
 
@@ -67,20 +68,15 @@ static void help(){
   exit(EXIT_SUCCESS);
 }
 static void parseCommandLineArgs(int argc, char** argv,char* videos[],char* gts[],
-        int* vc,char* algorithms[],char* initBoxes[][CMDLINEMAX],int* ac){
+        int* vc,char* algorithms[],char* initBoxes[][CMDLINEMAX],int* ac,char keys[CMDLINEMAX][LINEMAX]){
 
     *ac=*vc=0;
     for(int i=1;i<argc;i++){
         if(argv[i][0]=='-'){
-            char *key=(argv[i]+1),*argument=NULL;
-            if(std::strcmp("h",key)==0||std::strcmp("help",key)==0){
-                help();
+            for(int j=0;j<CMDLINEMAX;j++){
+                if(!strcmp(argv[i],keys[j])){
+                    keys[j][0]='\0';
                 }
-            if((argument=strchr(argv[i],'='))==NULL){
-                i++;
-                argument=argv[i];
-            }else{
-                argument++;
             }
             continue;
         }
@@ -193,6 +189,8 @@ static AssessmentRes assessment(char* video,char* gt_str, char* algorithms[],cha
   int linecount=0;
   Rect2d boundingBox;
   vector<double> averageMillisPerFrame(algnum,0.0);
+  static int videoNum=0;
+  videoNum++;
 
   FILE* gt=fopen(gt_str,"r");
   if(gt==NULL){
@@ -312,6 +310,11 @@ static AssessmentRes assessment(char* video,char* gt_str, char* algorithms[],cha
               res.results[i][j]->assess(boundingBox,initBoxes[i]);
       }
       imshow( "Tracking API", image );
+      if(saveImageKey){
+          char inbuf[LINEMAX];
+          sprintf(inbuf,"image%d_%d.jpg",videoNum,frameCounter);
+          imwrite(inbuf,image);
+      }
 
       if((frameCounter+1)>=ASSESS_TILL){
           break;
@@ -342,7 +345,11 @@ int main( int argc, char** argv ){
   palette.push_back(Scalar(0,255,255));
   int vcount=0,acount=0;
   char* videos[CMDLINEMAX],*gts[CMDLINEMAX],*algorithms[CMDLINEMAX],*initBoxes[CMDLINEMAX][CMDLINEMAX];
-  parseCommandLineArgs(argc,argv,videos,gts,&vcount,algorithms,initBoxes,&acount);
+  char keys[CMDLINEMAX][LINEMAX];
+  strcpy(keys[0],"-s");
+  parseCommandLineArgs(argc,argv,videos,gts,&vcount,algorithms,initBoxes,&acount,keys);
+  saveImageKey=(keys[0][0]=='\0');
+
   CV_Assert(acount<CMDLINEMAX && vcount<CMDLINEMAX);
   printf("videos and gts\n");
   for(int i=0;i<vcount;i++){
@@ -361,7 +368,7 @@ int main( int argc, char** argv ){
   for(int i=0;i<vcount;i++){
       results.push_back(assessment(videos[i],gts[i],algorithms,((char**)initBoxes)+i,acount));
   }
-  CV_Assert(results[0].results[0].size()<CMDLINEMAX);
+  CV_Assert( (int)results[0].results[0].size() < CMDLINEMAX );
   printf("\n\n");
 
   char buf[CMDLINEMAX*CMDLINEMAX*LINEMAX], buf2[CMDLINEMAX*40];

modules/tracking/src/TLD.hpp → modules/tracking/src/tld_tracker.hpp

@@ -42,8 +42,8 @@
 #include "precomp.hpp"
 #include "opencv2/video/tracking.hpp"
 #include "opencv2/imgproc.hpp"
-#include <algorithm>
-#include <limits.h>
+#include<algorithm>
+#include<limits.h>
 
 namespace cv {namespace tld
 {
@@ -57,53 +57,68 @@ namespace cv {namespace tld
 #define dfprintf(x)
 #define dprintf(x)
 #endif
-#define MEASURE_TIME(a) {\
-    clock_t start;float milisec=0.0;\
-    start=clock();{a} milisec=1000.0*(clock()-start)/CLOCKS_PER_SEC;\
-    dprintf(("%-90s took %f milis\n",#a,milisec)); }
-#define HERE dprintf(("%d\n",__LINE__));fflush(stderr);
-#define START_TICK(name) { clock_t start;double milisec=0.0; start=clock();
-#define END_TICK(name) milisec=1000.0*(clock()-start)/CLOCKS_PER_SEC;\
-    dprintf(("%s took %f milis\n",name,milisec)); }
+#define MEASURE_TIME(a)\
+{\
+    clock_t start; float milisec = 0.0; \
+    start = clock(); {a} milisec = 1000.0 * (clock() - start) / CLOCKS_PER_SEC; \
+    dprintf(("%-90s took %f milis\n", #a, milisec));\
+}
+#define HERE dprintf(("line %d\n", __LINE__)); fflush(stderr);
+#define START_TICK(name)\
+{ \
+    clock_t start; double milisec = 0.0; start = clock();
+#define END_TICK(name) milisec = 1000.0 * (clock() - start) / CLOCKS_PER_SEC; \
+    dprintf(("%s took %f milis\n", name, milisec)); \
+}
 extern Rect2d etalon;
 void myassert(const Mat& img);
 void printPatch(const Mat_<uchar>& standardPatch);
 std::string type2str(const Mat& mat);
-void drawWithRects(const Mat& img,std::vector<Rect2d>& blackOnes,Rect2d whiteOne=Rect2d(-1.0,-1.0,-1.0,-1.0));
-void drawWithRects(const Mat& img,std::vector<Rect2d>& blackOnes,std::vector<Rect2d>& whiteOnes);
+void drawWithRects(const Mat& img, std::vector<Rect2d>& blackOnes, Rect2d whiteOne = Rect2d(-1.0, -1.0, -1.0, -1.0));
+void drawWithRects(const Mat& img, std::vector<Rect2d>& blackOnes, std::vector<Rect2d>& whiteOnes, String fileName = "");
 
 //aux functions and variables
-//#define CLIP(x,a,b) MIN(MAX((x),(a)),(b))
-template<typename T> inline T CLIP(T x,T a,T b){return MIN(MAX(x,a),b);}
-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);
+template<typename T> inline T CLIP(T x, T a, T b){ return std::min(std::max(x, a), b); }
+/** Computes overlap between the two given rectangles. Overlap is computed as ratio of rectangles' intersection to that
+ * of their union.*/
+double overlap(const Rect2d& r1, const Rect2d& r2);
+/** Resamples the area surrounded by r2 in img so it matches the size of samples, where it is written.*/
+void resample(const Mat& img, const RotatedRect& r2, Mat_<uchar>& samples);
+/** Specialization of resample() for rectangles without retation for better performance and simplicity.*/
+void resample(const Mat& img, const Rect2d& r2, Mat_<uchar>& samples);
+/** Computes the variance of single given image.*/
 double variance(const Mat& img);
-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);
-unsigned int getMedian(const std::vector<unsigned int>& values, int size=-1);
+/** Computes normalized corellation coefficient between the two patches (they should be
+ * of the same size).*/
+double NCC(const Mat_<uchar>& patch1, const 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, double scaleStep);
+int getMedian(const std::vector<int>& values, int size = -1);
 
-class TLDEnsembleClassifier{
+class TLDEnsembleClassifier
+{
 public:
-    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();
+    static int makeClassifiers(Size size, int measurePerClassifier, int gridSize, std::vector<TLDEnsembleClassifier>& classifiers);
+    void integrate(const Mat_<uchar>& patch, bool isPositive);
+    double posteriorProbability(const uchar* data, int rowstep) const;
+    double posteriorProbabilityFast(const uchar* data) const;
+    void prepareClassifier(int rowstep);
 private:
-    static int getGridSize();
-    inline void stepPrefSuff(std::vector<uchar>& arr,int len);
-    void preinit(int ordinal);
-    unsigned short int code(const uchar* data,int rowstep)const;
-    std::vector<unsigned int> pos,neg;
-    std::vector<uchar> x1,y1,x2,y2;
+    TLDEnsembleClassifier(const std::vector<Vec4b>& meas, int beg, int end);
+    static void stepPrefSuff(std::vector<Vec4b> & arr, int pos, int len, int gridSize);
+    int code(const uchar* data, int rowstep) const;
+    int codeFast(const uchar* data) const;
+    std::vector<Point2i> posAndNeg;
+    std::vector<Vec4b> measurements;
+    std::vector<Point2i> offset;
+    int lastStep_;
 };
 
-class TrackerProxy{
+class TrackerProxy
+{
 public:
-    virtual bool init( const Mat& image, const Rect2d& boundingBox)=0;
-    virtual bool update(const Mat& image, Rect2d& boundingBox)=0;
+    virtual bool init(const Mat& image, const Rect2d& boundingBox) = 0;
+    virtual bool update(const Mat& image, Rect2d& boundingBox) = 0;
     virtual ~TrackerProxy(){}
 };