void Ftr::toViz( vecFtr &ftrs, const char *dirname )
{
char fname[1024];
Matrixu img;
for( uint k=0; k<ftrs.size(); k++ ){
sprintf_s(fname,"%s/ftr%05d.png",dirname,k);
img = ftrs[k]->toViz();
img.SaveImage(fname);
}
}
/****************************************************************
Tracker::ReplaysTracker
Replays the tracker based on the saved output.
Exceptions:
None
****************************************************************/
void Tracker::ReplayTrackers(vector<Matrixu> &vid, vector<string> statesfile, string outputvid, Matrixu colors)
{
Matrixu states;
vector<Matrixu> resvid(vid.size());
Matrixu colorframe;
// save videoList file
CvVideoWriter* w = NULL;
if ( ! outputvid.empty() )
{
w = cvCreateVideoWriter( outputvid.c_str(), CV_FOURCC('x','v','i','d'), 15, cvSize(vid[0].cols(), vid[0].rows()), 3 );
if ( w==NULL )
{
abortError(__LINE__,__FILE__,"Error opening videoList file for output");
}
}
for ( uint k=0; k<vid.size(); k++ )
{
vid[k].conv2RGB(resvid[k]);
resvid[k].drawText(("#"+int2str(k,3)).c_str(),1,25,255,255,0);
}
for ( uint j=0; j < statesfile.size(); j++ )
{
states.DLMRead(statesfile[j].c_str());
for( uint k=0; k<vid.size(); k++ )
{
resvid[k].drawRect(states(k,3),states(k,2),states(k,0),states(k,1),1,0,3,colors(j,0),colors(j,1),colors(j,2));
}
}
for ( uint k=0; k<vid.size(); k++ )
{
resvid[k]._keepIpl=true;
resvid[k].display(0,2);
cvWaitKey(1);
if( w!=NULL && k<vid.size()-1)
{
Matrixu::WriteFrame(w, resvid[k]);
}
resvid[k]._keepIpl=false; resvid[k].freeIpl();
}
// clean up
if( w != NULL )
{
cvReleaseVideoWriter( &w );
}
}
bool Tracker::initFace(TrackerParams* params, Matrixu &frame)
{
const char* cascade_name = "haarcascade_frontalface_alt_tree.xml";
const int minsz = 20;
if( Tracker::facecascade == NULL )
Tracker::facecascade = (CvHaarClassifierCascade*)cvLoad( cascade_name, 0, 0, 0 );
frame.createIpl();
IplImage *img = frame.getIpl();
IplImage* gray = cvCreateImage( cvSize(img->width, img->height), IPL_DEPTH_8U, 1 );
cvCvtColor(img, gray, CV_BGR2GRAY );
frame.freeIpl();
cvEqualizeHist(gray, gray);
CvMemStorage* storage = cvCreateMemStorage(0);
cvClearMemStorage(storage);
CvSeq* faces = cvHaarDetectObjects(gray, Tracker::facecascade, storage, 1.05, 3, CV_HAAR_DO_CANNY_PRUNING ,cvSize(minsz, minsz));
int index = faces->total-1;
CvRect* r = (CvRect*)cvGetSeqElem( faces, index );
while(r && (r->width<minsz || r->height<minsz || (r->y+r->height+10)>frame.rows() || (r->x+r->width)>frame.cols() ||
r->y<0 || r->x<0)){
r = (CvRect*)cvGetSeqElem( faces, --index);
}
//if( r == NULL ){
// cout << "ERROR: no face" << endl;
// return false;
//}
//else
// cout << "Face Found: " << r->x << " " << r->y << " " << r->width << " " << r->height << endl;
if( r==NULL )
return false;
//fprintf(stderr,"x=%f y=%f xmax=%f ymax=%f imgw=%f imgh=%f\n",(float)r->x,(float)r->y,(float)r->x+r->width,(float)r->y+r->height,(float)frame.cols(),(float)frame.rows());
params->_initstate.resize(4);
params->_initstate[0] = (float)r->x;// - r->width;
params->_initstate[1] = (float)r->y;// - r->height;
params->_initstate[2] = (float)r->width;
params->_initstate[3] = (float)r->height+10;
return true;
}
bool SimpleTracker::init(Matrixu frame, SimpleTrackerParams p, ClfStrongParams *clfparams)
{
static Matrixu *img;
img = &frame;
frame.initII();
_clf = ClfStrong::makeClf(clfparams);
_curState.resize(4);
for(int i=0;i<4;i++ ) _curState[i] = p._initstate[i];
SampleSet posx, negx;
fprintf(stderr,"Initializing Tracker..\n");
// sample positives and negatives from first frame
posx.sampleImage(img, (uint)_curState[0],(uint)_curState[1], (uint)_curState[2], (uint)_curState[3], p._init_postrainrad);
negx.sampleImage(img, (uint)_curState[0],(uint)_curState[1], (uint)_curState[2], (uint)_curState[3], 2.0f*p._srchwinsz, (1.5f*p._init_postrainrad), p._init_negnumtrain);
if( posx.size()<1 || negx.size()<1 ) return false;
// train
_clf->update(posx,negx);
negx.clear();
img->FreeII();
_trparams = p;
_clfparams = clfparams;
_cnt = 0;
return true;
}
void Tracker::replayTracker(vector<Matrixu> &vid, string statesfile, string outputvid, uint R, uint G, uint B)
{
Matrixf states;
states.DLMRead(statesfile.c_str());
Matrixu colorframe;
// save video file
CvVideoWriter* w = NULL;
if( ! outputvid.empty() ){
w = cvCreateVideoWriter( outputvid.c_str(), CV_FOURCC('I','Y','U','V'), 15, cvSize(vid[0].cols(), vid[0].rows()), 3 );
if( w==NULL ) abortError(__LINE__,__FILE__,"Error opening video file for output");
}
for( uint k=0; k<vid.size(); k++ )
{
vid[k].conv2RGB(colorframe);
colorframe.drawRect(states(k,2),states(k,3),states(k,0),states(k,1),1,0,2,R,G,B);
colorframe.drawText(("#"+int2str(k,3)).c_str(),1,25,255,255,0);
colorframe._keepIpl=true;
colorframe.display(1,2);
cvWaitKey(1);
if( w != NULL )
cvWriteFrame( w, colorframe.getIpl() );
colorframe._keepIpl=false; colorframe.freeIpl();
}
// clean up
if( w != NULL )
cvReleaseVideoWriter( &w );
}
void SimpleTracker::track_frames(vector<Matrixu> &video, SimpleTrackerParams p, ClfStrongParams *clfparams)
{
CvVideoWriter* w = NULL;
Matrixu states(video.size(), 4);
Matrixu t; string pnd="#";
// save video file
if( ! p._vidsave.empty() ){
w = cvCreateVideoWriter( p._vidsave.c_str(), CV_FOURCC('I','Y','U','V'), 15, cvSize(video[0].cols(), video[0].rows()), 3 );
if( w==NULL ) abortError(__LINE__,__FILE__,"Error opening video file for output");
}
// initialization
int frameind = 0;
if( p._initWithFace ){ // init with face
fprintf(stderr,"Searching for face...\n");
while( !Tracker::initFace(&p,video[frameind]) ){
video[frameind].conv2RGB(t); t._keepIpl=true;
t.drawText(("#"+int2str(frameind,3)).c_str(),1,25,255,255,0);
// display on screen & write to disk
if( p._disp ){ t.display(1); cvWaitKey(1); }
if( w != NULL ) cvWriteFrame( w, t.getIpl() );
t._keepIpl=false; t.freeIpl();
//frameind++;
}
clfparams->_ftrParams->_width = (uint)p._initstate[2];
clfparams->_ftrParams->_height = (uint)p._initstate[3];
init(video[frameind], p, clfparams);
} // init with params
else{
clfparams->_ftrParams->_width = (uint)p._initstate[2];
clfparams->_ftrParams->_height = (uint)p._initstate[3];
init(video[0], p, clfparams);
states(frameind,0) = (uint)_curState[0];
states(frameind,1) = (uint)_curState[1];
states(frameind,2) = (uint)_curState[2];
states(frameind,3) = (uint)_curState[3];
//frameind++;
}
// track rest of frames
StopWatch sw(true); double ttt;
for( frameind; frameind<(int)video.size(); frameind++ )
{
ttt = sw.Elapsed(true);
fprintf(stderr,"%s%d Frames/%f sec = %f FPS",ERASELINE,frameind,ttt,((double)frameind)/ttt);
track_frame(video[frameind],t);
if( p._disp || w!= NULL ){
video[frameind] = t;
video[frameind].drawText(("#"+int2str(frameind,3)).c_str(),1,25,255,255,0);
video[frameind].createIpl();
video[frameind]._keepIpl = true;
// display on screen
if( p._disp ){
video[frameind].display(1);
cvWaitKey(1);
}
// save video
if( w != NULL && frameind<(int)video.size()-1 )
Matrixu::WriteFrame(w, video[frameind]);
video[frameind]._keepIpl = false;
video[frameind].freeIpl();
}
for( int s=0; s<4; s++ ) states(frameind,s) = (uint)_curState[s];
}
// save states
if( !p._trsave.empty() ){
bool scs = states.DLMWrite(p._trsave.c_str());
if( !scs ) abortError(__LINE__,__FILE__,"error saving states to file");
}
// clean up
if( w != NULL )
cvReleaseVideoWriter( &w );
}
double SimpleTracker::track_frame(Matrixu &frame, Matrixu &framedisp)
{
static SampleSet posx, negx, detectx;
static vectorf prob;
static vectori order;
static Matrixu *img;
double resp;
// copy a color version into framedisp (this is where we will draw a colored box around the object for output)
frame.conv2RGB(framedisp);
img = &frame;
frame.initII();
// run current clf on search window
detectx.sampleImage(img,(uint)_curState[0],(uint)_curState[1],(uint)_curState[2],(uint)_curState[3], (float)_trparams._srchwinsz);
prob = _clf->classify(detectx,_trparams._useLogR);
/////// DEBUG /////// display actual probability map
if( _trparams._debugv ){
Matrixf probimg(frame.rows(),frame.cols());
for( uint k=0; k<(uint)detectx.size(); k++ )
probimg(detectx[k]._row, detectx[k]._col) = prob[k];
probimg.convert2img().display(2,2);
cvWaitKey(1);
}
// find best location
int bestind = max_idx(prob);
resp=prob[bestind];
_curState[1] = (float)detectx[bestind]._row;
_curState[0] = (float)detectx[bestind]._col;
// train location clf (negx are randomly selected from image, posx is just the current tracker location)
if( _trparams._negsamplestrat == 0 )
negx.sampleImage(img, _trparams._negnumtrain, (int)_curState[2], (int)_curState[3]);
else
negx.sampleImage(img, (int)_curState[0], (int)_curState[1], (int)_curState[2], (int)_curState[3],
(1.5f*_trparams._srchwinsz), _trparams._posradtrain+5, _trparams._negnumtrain);
if( _trparams._posradtrain == 1 )
posx.push_back(img, (int)_curState[0], (int)_curState[1], (int)_curState[2], (int)_curState[3]);
else
posx.sampleImage(img, (int)_curState[0], (int)_curState[1], (int)_curState[2], (int)_curState[3], _trparams._posradtrain, 0, _trparams._posmaxtrain);
_clf->update(posx,negx);
/////// DEBUG /////// display sampled negative points
if( _trparams._debugv ){
for( int j=0; j<negx.size(); j++ )
framedisp.drawEllipse(1,1,(float)negx[j]._col,(float)negx[j]._row,1,255,0,255);
}
// clean up
img->FreeII();
posx.clear(); negx.clear(); detectx.clear();
// draw a colored box around object
framedisp.drawRect(_curState[2], _curState[3], _curState[0], _curState[1], 1, 0,
_trparams._lineWidth, _trparams._boxcolor[0], _trparams._boxcolor[1], _trparams._boxcolor[2] );
_cnt++;
return resp;
}
/****************************************************************
Tracker::InitializeWithFace
Initialize the tracker with opencv's face detector.
Exceptions:
None
****************************************************************/
bool Tracker::InitializeWithFace( TrackerParameters* params, Matrixu& frame )
{
const int minsz = 20;
//Get the name of the haar-cascade .xml file
const char* cascade_name = HAAR_CASCADE_FILE_NAME;
ASSERT_TRUE( cascade_name != NULL );
//Load the cascade
if ( Tracker::s_faceCascade == NULL )
{
Tracker::s_faceCascade = (CvHaarClassifierCascade*)cvLoad( cascade_name, 0, 0, 0 );
}
frame.createIpl();
IplImage* img = frame.getIpl();
ASSERT_TRUE( img != NULL );
//convert to grayscale
IplImage* gray = cvCreateImage( cvSize(img->width, img->height), IPL_DEPTH_8U, 1 );
ASSERT_TRUE( gray != NULL );
cvCvtColor(img, gray, CV_BGR2GRAY );
frame.freeIpl();
//histogram equalization
cvEqualizeHist( gray, gray );
//memory storage
CvMemStorage* storage = cvCreateMemStorage(0);
cvClearMemStorage(storage);
//call opencv's haar feature based detector
CvSeq* faces = cvHaarDetectObjects( gray,
Tracker::s_faceCascade,
storage,
1.05,
3,
CV_HAAR_DO_CANNY_PRUNING,
cvSize( minsz, minsz ) );
int index = faces->total-1;
CvRect* r = (CvRect*)cvGetSeqElem( faces, index );
if ( r == NULL )
{
return false;
}
while ( r && (r->width<minsz || r->height < minsz || (r->y+r->height+10)>frame.rows() || (r->x+r->width)>frame.cols() ||
r->y<0 || r->x<0) )
{
r = (CvRect*)cvGetSeqElem( faces, --index );
}
//set the params
params->m_initState.resize(4);
params->m_initState[0] = (float)r->x;
params->m_initState[1] = (float)r->y;
params->m_initState[2] = (float)r->width;
params->m_initState[3] = (float)r->height+10;
return true;
}