MIL, Boosting tracker test in opencv 3.0

I have tested MIL, Boosting tracking algorithm in opencv 3.0

Please refer to official reference here -> http://docs.opencv.org/trunk/modules/tracking/doc/tracking.html

Firstly, to use tracker algorithm, you have to set the value which is tracking module path of OPENCV_EXTRA_MODUALES_PATH option, when you build the opencv 3.0.



We can download tracking module in Github. -> https://github.com/Itseez/opencv_contrib
"opencv_contrib" is developing separately with opencv full version.
In setting cmake, I checked only tracking option. because other module may be unstable version yet.
And I need just tracking module.


The example source code is referenced in here -> https://github.com/lenlen/opencv/blob/tracking_api/samples/cpp/tracker.cpp

In source code, you can select Mil and Boosting algorithm option in the function of creation.

At the result, MIL is faster than boosting.
But both algorithm are very slow and performance is also not good yet.
So they can not use in industrial application.


MIL result video

Boosting result video


The example source code is here





#include < stdio.h>  
#include < opencv2\video\tracker.hpp>
#include < opencv2\opencv.hpp>


 
#ifdef _DEBUG          
#pragma comment(lib, "opencv_core300d.lib")  
#pragma comment(lib, "opencv_highgui300d.lib")  
#pragma comment(lib, "opencv_imgcodecs300d.lib")  
#pragma comment(lib, "opencv_videoio300d.lib") 
#pragma comment(lib, "opencv_imgproc300d.lib") 
#pragma comment(lib, "opencv_cuda300d.lib")   
#pragma comment(lib, "opencv_cudawarping300d.lib")
#pragma comment(lib, "opencv_tracking300d.lib")
#define _DEBUG_RRRR  
#else  
#pragma comment(lib, "opencv_core300.lib")  
#pragma comment(lib, "opencv_highgui300.lib")  
#pragma comment(lib, "opencv_imgcodecs300.lib")  
#pragma comment(lib, "opencv_videoio300.lib") 
#pragma comment(lib, "opencv_imgproc300.lib") 
#pragma comment(lib, "opencv_cuda300.lib")   
#pragma comment(lib, "opencv_cudawarping300.lib")
#pragma comment(lib, "opencv_tracking300d.lib")
#endif  


using namespace cv;
using namespace std;

static Mat image;
static bool selectObject = false;
static bool startSelection = false;
static Rect2d boundingBox;
static bool paused;

static void onMouse( int event, int x, int y, int, void* ) 
{ 
 if( !selectObject ) 
 { 
  switch ( event ) 
  { 
  case EVENT_LBUTTONDOWN: 
   //set origin of the bounding box 
   startSelection = true; 
   boundingBox.x = x; 
   boundingBox.y = y; 
   break; 
  case EVENT_LBUTTONUP: 
   //sei with and height of the bounding box 
   boundingBox.width = std::abs( x - boundingBox.x ); 
   boundingBox.height = std::abs( y - boundingBox.y ); 
   paused = false; 
   selectObject = true; 

   printf("Object Rect Size(left, right, width, height) %.1f %.1f %.1f %.1f\n", boundingBox.x, boundingBox.y, boundingBox.width, boundingBox.height);
   break; 
  case EVENT_MOUSEMOVE: 
   if( startSelection && !selectObject ) 
   { 
    //draw the bounding box 
    Mat currentFrame; 
    image.copyTo( currentFrame ); 
    rectangle( currentFrame, Point( boundingBox.x, boundingBox.y ), Point( x, y ), Scalar( 255, 0, 0 ), 2, 1 ); 
    imshow( "Tracking API", currentFrame ); 
   } 
   break; 
  } 
 } 
} 




int main()
{
 unsigned long AAtime=0, BBtime=0;

 VideoCapture cap;
 cap.open("C:\\videoSample\\tracking\\sample2.avi");

 if( !cap.isOpened() ) 
 { 
  printf("Video File Open Fail! \n");
  return -1; 
 } 


 Mat frame; 
 paused = false; 
 namedWindow( "Tracking", 0 ); 
 setMouseCallback( "Tracking", onMouse, 0 );


 //instantiates the specific Tracker 
 //MIL : TrackerMIL
 //BOOSTING : TrackerBoosting
 Ptr< Tracker> tracker = Tracker::create("BOOSTING" ); //"MIL");//
 if( tracker == NULL ) 
 { 
  printf("Error in the instantiation of the tracker..\n");
  return -1; 
 } 

 //get the first frame 
 cap >> frame; 
 frame.copyTo( image ); 
 imshow( "Tracking", image ); 

 bool initialized = false;
 while(1)
 { 
  if( !paused ) 
  { 
   cap >> frame; 
   frame.copyTo( image ); 


   if( !initialized && selectObject ) 
   { 
    //initializes the tracker 
    AAtime = getTickCount();
    if( !tracker->init( frame, boundingBox ) ) 
    { 
     printf("Could not initialize tracker\n"); 
     return -1; 
    } 
    BBtime = getTickCount();
    double fps = (BBtime - AAtime)/getTickFrequency();
    double sec = 1/fps;
    printf("Tracking Initial time = %.2lffsp, %.2lfsec \n",  fps, sec );

    initialized = true; 
   } 
   else if( initialized ) 
   { 
    AAtime = getTickCount();
    
    //updates the tracker 
    if( tracker->update( frame, boundingBox ) ) 
    { 
     rectangle( image, boundingBox, Scalar( 255, 0, 0 ), 2, 1 ); 
    } 

    BBtime = getTickCount();
    double fps = (BBtime - AAtime)/getTickFrequency();
    double sec = 1/fps;
    printf("Tracking update time = %.2lffsp, %.2lfsec \n",  fps, sec );

   } 
   imshow( "Tracking", image ); 
  } 


  char c = (char) waitKey( 2 ); 
  if( c == 'q' ) 
   break; 
  if( c == 'p' ) 
   paused = !paused; 


 } 
}

Comments

  1. Hi there.
    Thanks for your code and it helped me a lot with my home work. According to my experiments, at first, the fps rate is as low as what you show in Youtube. However, then I realized that it's because I ran under debug mode. As soon as I switch to release mode, fps reached 50~60. Check it out please.
    In addition, this paper is where the tracker API originates from. It says in 2006 they could reach 20 fps. That's why I suspected the results at first.
    Here is the link:
    http://www.vision.ee.ethz.ch/~hegrabne/papers/Grabner2006Real-TimeTrackingvia.pdf

    More questions, contact me at ae86208@gmail.com .
    Best regards.

    ReplyDelete
    Replies
    1. Refer to this video
      https://www.youtube.com/watch?v=1oBWl9a8SKo&feature=youtu.be

      Delete

Post a Comment

Popular posts from this blog

OpenCV Stitching example (Stitcher class, Panorama)

(OpenCV Study) Background subtractor MOG, MOG2, GMG example source code (BackgroundSubtractorMOG, BackgroundSubtractorMOG2, BackgroundSubtractorGMG)

Example source code of extract HOG feature from images, save descriptor values to xml file, using opencv (using HOGDescriptor )

Real-time N camera stitching Class.

8 point algorithm (Matlab source code) / The method to get the Fundamental Matrix and the Essential matrix

Optical Flow sample source code using OpenCV

Video Stabilization example source code, (using cvFindHomography, cvWarpPerspective functions in openCV)

(OpenCV Study) calcOpticalFlowFarneback example source code ( dense optical flow )

yuv422(YUYV) to RGB and RGB to yuv422(YUYV), (Using OpenCV and TBB)

OpenCV Drawing Example, (line, circle, rectangle, ellipse, polyline, fillConvexPoly, putText, drawContours)