mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-12-12 15:19:11 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity
d122510c4f
opencv/modules/legacy/src/blobtrackingmsfg.cpp
Roman Donchenko e9a28f66ee Normalized file endings.
2013-08-21 18:59:25 +04:00

1182 lines
38 KiB
C++
Raw Blame History

/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// Intel License Agreement
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#ifdef _OPENMP
#include "omp.h"
#endif
// Uncomment to trade flexibility for speed
//#define CONST_HIST_SIZE
// Uncomment to get some performance stats in stderr
//#define REPORT_TICKS
#ifdef CONST_HIST_SIZE
#define m_BinBit 5
#define m_ByteShift 3
#endif
typedef float DefHistType;
#define DefHistTypeMat CV_32F
#define HIST_INDEX(_pData) (((_pData)[0]>>m_ByteShift) + (((_pData)[1]>>(m_ByteShift))<<m_BinBit)+((pImgData[2]>>m_ByteShift)<<(m_BinBit*2)))
class DefHist
{
public:
CvMat* m_pHist;
DefHistType m_HistVolume;
DefHist(int BinNum=0)
{
m_pHist = NULL;
m_HistVolume = 0;
Resize(BinNum);
}
~DefHist()
{
if(m_pHist)cvReleaseMat(&m_pHist);
}
void Resize(int BinNum)
{
if(m_pHist)cvReleaseMat(&m_pHist);
if(BinNum>0)
{
m_pHist = cvCreateMat(1, BinNum, DefHistTypeMat);
cvZero(m_pHist);
}
m_HistVolume = 0;
}
void Update(DefHist* pH, float W)
{ /* Update histogram: */
double Vol, WM, WC;
Vol = 0.5*(m_HistVolume + pH->m_HistVolume);
WM = Vol*(1-W)/m_HistVolume;
WC = Vol*(W)/pH->m_HistVolume;
cvAddWeighted(m_pHist, WM, pH->m_pHist,WC,0,m_pHist);
m_HistVolume = (float)cvSum(m_pHist).val[0];
} /* Update histogram: */
};
class CvBlobTrackerOneMSFG:public CvBlobTrackerOne
{
protected:
int m_BinNumTotal; /* protected only for parralel MSPF */
CvSize m_ObjSize;
void ReAllocKernel(int w, int h)
{
int x,y;
float x0 = 0.5f*(w-1);
float y0 = 0.5f*(h-1);
assert(w>0);
assert(h>0);
m_ObjSize = cvSize(w,h);
if(m_KernelHist) cvReleaseMat(&m_KernelHist);
if(m_KernelMeanShift) cvReleaseMat(&m_KernelMeanShift);
m_KernelHist = cvCreateMat(h, w, DefHistTypeMat);
m_KernelMeanShift = cvCreateMat(h, w, DefHistTypeMat);
for(y=0; y<h; ++y) for(x=0; x<w; ++x)
{
double r2 = ((x-x0)*(x-x0)/(x0*x0)+(y-y0)*(y-y0)/(y0*y0));
// double r2 = ((x-x0)*(x-x0)+(y-y0)*(y-y0))/((y0*y0)+(x0*x0));
CV_MAT_ELEM(m_KernelHist[0],DefHistType, y, x) = (DefHistType)GetKernelHist(r2);
CV_MAT_ELEM(m_KernelMeanShift[0],DefHistType, y, x) = (DefHistType)GetKernelMeanShift(r2);
}
}
private:
/* Parameters: */
int m_IterNum;
float m_FGWeight;
float m_Alpha;
CvMat* m_KernelHist;
CvMat* m_KernelMeanShift;
#ifndef CONST_HIST_SIZE
int m_BinBit;
int m_ByteShift;
#endif
int m_BinNum;
int m_Dim;
/*
CvMat* m_HistModel;
float m_HistModelVolume;
CvMat* m_HistCandidate;
float m_HistCandidateVolume;
CvMat* m_HistTemp;
*/
DefHist m_HistModel;
DefHist m_HistCandidate;
DefHist m_HistTemp;
CvBlob m_Blob;
int m_Collision;
void ReAllocHist(int Dim, int BinBit)
{
#ifndef CONST_HIST_SIZE
m_BinBit = BinBit;
m_ByteShift = 8-BinBit;
#endif
m_Dim = Dim;
m_BinNum = (1<<BinBit);
m_BinNumTotal = cvRound(pow((double)m_BinNum,(double)m_Dim));
/*
if(m_HistModel) cvReleaseMat(&m_HistModel);
if(m_HistCandidate) cvReleaseMat(&m_HistCandidate);
if(m_HistTemp) cvReleaseMat(&m_HistTemp);
m_HistCandidate = cvCreateMat(1, m_BinNumTotal, DefHistTypeMat);
m_HistModel = cvCreateMat(1, m_BinNumTotal, DefHistTypeMat);
m_HistTemp = cvCreateMat(1, m_BinNumTotal, DefHistTypeMat);
cvZero(m_HistCandidate);
cvZero(m_HistModel);
m_HistModelVolume = 0.0f;
m_HistCandidateVolume = 0.0f;
*/
m_HistCandidate.Resize(m_BinNumTotal);
m_HistModel.Resize(m_BinNumTotal);
m_HistTemp.Resize(m_BinNumTotal);
}
double GetKernelHist(double r2)
{
return (r2 < 1) ? 1 - r2 : 0;
}
double GetKernelMeanShift(double r2)
{
return (r2<1) ? 1 : 0;
}
// void CollectHist(IplImage* pImg, IplImage* pMask, CvPoint Center, CvMat* pHist, DefHistType* pHistVolume)
// void CollectHist(IplImage* pImg, IplImage* pMask, CvPoint Center, DefHist* pHist)
void CollectHist(IplImage* pImg, IplImage* pMask, CvBlob* pBlob, DefHist* pHist)
{
int UsePrecalculatedKernel = 0;
int BW = cvRound(pBlob->w);
int BH = cvRound(pBlob->h);
DefHistType Volume = 0;
int x0 = cvRound(pBlob->x - BW*0.5);
int y0 = cvRound(pBlob->y - BH*0.5);
int x,y;
UsePrecalculatedKernel = (BW == m_ObjSize.width && BH == m_ObjSize.height ) ;
//cvZero(pHist);
cvSet(pHist->m_pHist,cvScalar(1.0/m_BinNumTotal)); /* no zero bins, all bins have very small value*/
Volume = 1;
assert(BW < pImg->width);
assert(BH < pImg->height);
if((x0+BW)>=pImg->width) BW=pImg->width-x0-1;
if((y0+BH)>=pImg->height) BH=pImg->height-y0-1;
if(x0<0){ x0=0;}
if(y0<0){ y0=0;}
if(m_Dim == 3)
{
for(y=0; y<BH; ++y)
{
unsigned char* pImgData = &CV_IMAGE_ELEM(pImg,unsigned char,y+y0,x0*3);
unsigned char* pMaskData = pMask?(&CV_IMAGE_ELEM(pMask,unsigned char,y+y0,x0)):NULL;
DefHistType* pKernelData = NULL;
if(UsePrecalculatedKernel)
{
pKernelData = ((DefHistType*)CV_MAT_ELEM_PTR_FAST(m_KernelHist[0],y,0,sizeof(DefHistType)));
}
for(x=0; x<BW; ++x, pImgData+=3)
{
DefHistType K;
int index = HIST_INDEX(pImgData);
assert(index >= 0 && index < pHist->m_pHist->cols);
if(UsePrecalculatedKernel)
{
K = pKernelData[x];
}
else
{
float dx = (x+x0-pBlob->x)/(pBlob->w*0.5f);
float dy = (y+y0-pBlob->y)/(pBlob->h*0.5f);
double r2 = dx*dx+dy*dy;
K = (float)GetKernelHist(r2);
}
if(pMaskData)
{
K *= pMaskData[x]*0.003921568627450980392156862745098f;
}
Volume += K;
((DefHistType*)(pHist->m_pHist->data.ptr))[index] += K;
} /* Next column. */
} /* Next row. */
} /* if m_Dim == 3. */
pHist->m_HistVolume = Volume;
}; /* CollectHist */
double calcBhattacharyya(DefHist* pHM = NULL, DefHist* pHC = NULL, DefHist* pHT = NULL)
{
if(pHM==NULL) pHM = &m_HistModel;
if(pHC==NULL) pHC = &m_HistCandidate;
if(pHT==NULL) pHT = &m_HistTemp;
if(pHC->m_HistVolume*pHM->m_HistVolume > 0)
{
#if 0
// Use CV functions:
cvMul(pHM->m_pHist,pHC->m_pHist,pHT->m_pHist);
cvPow(pHT->m_pHist,pHT->m_pHist,0.5);
return cvSum(pHT->m_pHist).val[0] / sqrt(pHC->m_HistVolume*pHM->m_HistVolume);
#else
// Do computations manually and let autovectorizer do the job:
DefHistType* hm=(DefHistType *)(pHM->m_pHist->data.ptr);
DefHistType* hc=(DefHistType *)(pHC->m_pHist->data.ptr);
//ht=(DefHistType *)(pHT->m_pHist->data.ptr);
int size = pHM->m_pHist->width*pHM->m_pHist->height;
double sum = 0.;
for(int i = 0; i < size; i++ )
{
sum += sqrt(hm[i]*hc[i]);
}
return sum / sqrt(pHC->m_HistVolume*pHM->m_HistVolume);
#endif
}
return 0;
} /* calcBhattacharyyaCoefficient. */
protected:
// double GetBhattacharyya(IplImage* pImg, IplImage* pImgFG, float x, float y, DefHist* pHist=NULL)
double GetBhattacharyya(IplImage* pImg, IplImage* pImgFG, CvBlob* pBlob, DefHist* pHist=NULL, int /*thread_number*/ = 0)
{
if(pHist==NULL)pHist = &m_HistTemp;
CollectHist(pImg, pImgFG, pBlob, pHist);
return calcBhattacharyya(&m_HistModel, pHist, pHist);
}
void UpdateModelHist(IplImage* pImg, IplImage* pImgFG, CvBlob* pBlob)
{
if(m_Alpha>0 && !m_Collision)
{ /* Update histogram: */
CollectHist(pImg, pImgFG, pBlob, &m_HistCandidate);
m_HistModel.Update(&m_HistCandidate, m_Alpha);
} /* Update histogram. */
} /* UpdateModelHist */
public:
CvBlobTrackerOneMSFG()
{
/* Add several parameters for external use: */
m_FGWeight = 2;
AddParam("FGWeight", &m_FGWeight);
CommentParam("FGWeight","Weight of FG mask using (0 - mask will not be used for tracking)");
m_Alpha = 0.01f;
AddParam("Alpha", &m_Alpha);
CommentParam("Alpha","Coefficient for model histogram updating (0 - hist is not upated)");
m_IterNum = 10;
AddParam("IterNum", &m_IterNum);
CommentParam("IterNum","Maximal number of iteration in meanshift operation");
/* Initialize internal data: */
m_Collision = 0;
// m_BinBit=0;
m_Dim = 0;
/*
m_HistModel = NULL;
m_HistCandidate = NULL;
m_HistTemp = NULL;
*/
m_KernelHist = NULL;
m_KernelMeanShift = NULL;
ReAllocHist(3,5); /* 3D hist, each dim has 2^5 bins*/
SetModuleName("MSFG");
}
~CvBlobTrackerOneMSFG()
{
/*
if(m_HistModel) cvReleaseMat(&m_HistModel);
if(m_HistCandidate) cvReleaseMat(&m_HistCandidate);
if(m_HistTemp) cvReleaseMat(&m_HistTemp);
*/
if(m_KernelHist) cvReleaseMat(&m_KernelHist);
if(m_KernelMeanShift) cvReleaseMat(&m_KernelMeanShift);
}
/* Interface: */
virtual void Init(CvBlob* pBlobInit, IplImage* pImg, IplImage* pImgFG = NULL)
{
int w = cvRound(CV_BLOB_WX(pBlobInit));
int h = cvRound(CV_BLOB_WY(pBlobInit));
if(w<CV_BLOB_MINW)w=CV_BLOB_MINW;
if(h<CV_BLOB_MINH)h=CV_BLOB_MINH;
if(pImg)
{
if(w>pImg->width)w=pImg->width;
if(h>pImg->height)h=pImg->height;
}
ReAllocKernel(w,h);
if(pImg)
CollectHist(pImg, pImgFG, pBlobInit, &m_HistModel);
m_Blob = pBlobInit[0];
};
virtual CvBlob* Process(CvBlob* pBlobPrev, IplImage* pImg, IplImage* pImgFG = NULL)
{
int iter;
if(pBlobPrev)
{
m_Blob = pBlobPrev[0];
}
{ /* Check blob size and realloc kernels if it is necessary: */
int w = cvRound(m_Blob.w);
int h = cvRound(m_Blob.h);
if( w != m_ObjSize.width || h!=m_ObjSize.height)
{
ReAllocKernel(w,h);
/* after this ( w != m_ObjSize.width || h!=m_ObjSize.height) shoiuld be false */
}
} /* Check blob size and realloc kernels if it is necessary: */
for(iter=0; iter<m_IterNum; ++iter)
{
float newx=0,newy=0,sum=0;
//int x,y;
double B0;
//CvPoint Center = cvPoint(cvRound(m_Blob.x),cvRound(m_Blob.y));
CollectHist(pImg, NULL, &m_Blob, &m_HistCandidate);
B0 = calcBhattacharyya();
if(m_Wnd)if(CV_BLOB_ID(pBlobPrev)==0 && iter == 0)
{ /* Debug: */
IplImage* pW = cvCloneImage(pImgFG);
IplImage* pWFG = cvCloneImage(pImgFG);
int x,y;
cvZero(pW);
cvZero(pWFG);
assert(m_ObjSize.width < pImg->width);
assert(m_ObjSize.height < pImg->height);
/* Calculate shift vector: */
for(y=0; y<pImg->height; ++y)
{
unsigned char* pImgData = &CV_IMAGE_ELEM(pImg,unsigned char,y,0);
unsigned char* pMaskData = pImgFG?(&CV_IMAGE_ELEM(pImgFG,unsigned char,y,0)):NULL;
for(x=0; x<pImg->width; ++x, pImgData+=3)
{
int xk = cvRound(x-(m_Blob.x-m_Blob.w*0.5));
int yk = cvRound(y-(m_Blob.y-m_Blob.h*0.5));
double HM = 0;
double HC = 0;
double K;
int index = HIST_INDEX(pImgData);
assert(index >= 0 && index < m_BinNumTotal);
if(fabs(x-m_Blob.x)>m_Blob.w*0.6) continue;
if(fabs(y-m_Blob.y)>m_Blob.h*0.6) continue;
if(xk < 0 || xk >= m_KernelMeanShift->cols) continue;
if(yk < 0 || yk >= m_KernelMeanShift->rows) continue;
if(m_HistModel.m_HistVolume>0)
HM = ((DefHistType*)m_HistModel.m_pHist->data.ptr)[index]/m_HistModel.m_HistVolume;
if(m_HistCandidate.m_HistVolume>0)
HC = ((DefHistType*)m_HistCandidate.m_pHist->data.ptr)[index]/m_HistCandidate.m_HistVolume;
K = *(DefHistType*)CV_MAT_ELEM_PTR_FAST(m_KernelMeanShift[0],yk,xk,sizeof(DefHistType));
if(HC>0)
{
double V = sqrt(HM / HC);
int Vi = cvRound(V * 64);
if(Vi < 0) Vi = 0;
if(Vi > 255) Vi = 255;
CV_IMAGE_ELEM(pW,uchar,y,x) = (uchar)Vi;
V += m_FGWeight*(pMaskData?(pMaskData[x]/255.0f):0);
V*=K;
Vi = cvRound(V * 64);
if(Vi < 0) Vi = 0;
if(Vi > 255) Vi = 255;
CV_IMAGE_ELEM(pWFG,uchar,y,x) = (uchar)Vi;
}
} /* Next column. */
} /* Next row. */
//cvNamedWindow("MSFG_W",0);
//cvShowImage("MSFG_W",pW);
//cvNamedWindow("MSFG_WFG",0);
//cvShowImage("MSFG_WFG",pWFG);
//cvNamedWindow("MSFG_FG",0);
//cvShowImage("MSFG_FG",pImgFG);
//cvSaveImage("MSFG_W.bmp",pW);
//cvSaveImage("MSFG_WFG.bmp",pWFG);
//cvSaveImage("MSFG_FG.bmp",pImgFG);
} /* Debug. */
/* Calculate new position by meanshift: */
/* Calculate new position: */
if(m_Dim == 3)
{
int x0 = cvRound(m_Blob.x - m_ObjSize.width*0.5);
int y0 = cvRound(m_Blob.y - m_ObjSize.height*0.5);
int x,y;
assert(m_ObjSize.width < pImg->width);
assert(m_ObjSize.height < pImg->height);
/* Crop blob bounds: */
if((x0+m_ObjSize.width)>=pImg->width) x0=pImg->width-m_ObjSize.width-1;
if((y0+m_ObjSize.height)>=pImg->height) y0=pImg->height-m_ObjSize.height-1;
if(x0<0){ x0=0;}
if(y0<0){ y0=0;}
/* Calculate shift vector: */
for(y=0; y<m_ObjSize.height; ++y)
{
unsigned char* pImgData = &CV_IMAGE_ELEM(pImg,unsigned char,y+y0,x0*3);
unsigned char* pMaskData = pImgFG?(&CV_IMAGE_ELEM(pImgFG,unsigned char,y+y0,x0)):NULL;
DefHistType* pKernelData = (DefHistType*)CV_MAT_ELEM_PTR_FAST(m_KernelMeanShift[0],y,0,sizeof(DefHistType));
for(x=0; x<m_ObjSize.width; ++x, pImgData+=3)
{
DefHistType K = pKernelData[x];
double HM = 0;
double HC = 0;
int index = HIST_INDEX(pImgData);
assert(index >= 0 && index < m_BinNumTotal);
if(m_HistModel.m_HistVolume>0)
HM = ((DefHistType*)m_HistModel.m_pHist->data.ptr)[index]/m_HistModel.m_HistVolume;
if(m_HistCandidate.m_HistVolume>0)
HC = ((DefHistType*)m_HistCandidate.m_pHist->data.ptr)[index]/m_HistCandidate.m_HistVolume;
if(HC>0)
{
double V = sqrt(HM / HC);
if(!m_Collision && m_FGWeight>0 && pMaskData)
{
V += m_FGWeight*(pMaskData[x]/255.0f);
}
K *= (float)MIN(V,100000.);
}
sum += K;
newx += K*x;
newy += K*y;
} /* Next column. */
} /* Next row. */
if(sum > 0)
{
newx /= sum;
newy /= sum;
}
newx += x0;
newy += y0;
} /* if m_Dim == 3. */
/* Calculate new position by meanshift: */
for(;;)
{ /* Iterate using bahattcharrya coefficient: */
double B1;
CvBlob B = m_Blob;
// CvPoint NewCenter = cvPoint(cvRound(newx),cvRound(newy));
B.x = newx;
B.y = newy;
CollectHist(pImg, NULL, &B, &m_HistCandidate);
B1 = calcBhattacharyya();
if(B1 > B0) break;
newx = 0.5f*(newx+m_Blob.x);
newy = 0.5f*(newy+m_Blob.y);
if(fabs(newx-m_Blob.x)<0.1 && fabs(newy-m_Blob.y)<0.1) break;
} /* Iterate using bahattcharrya coefficient. */
if(fabs(newx-m_Blob.x)<0.5 && fabs(newy-m_Blob.y)<0.5) break;
m_Blob.x = newx;
m_Blob.y = newy;
} /* Next iteration. */
while(!m_Collision && m_FGWeight>0)
{ /* Update size if no collision. */
float Alpha = 0.04f;
CvBlob NewBlob;
double M00,X,Y,XX,YY;
CvMoments m;
CvRect r;
CvMat mat;
r.width = cvRound(m_Blob.w*1.5+0.5);
r.height = cvRound(m_Blob.h*1.5+0.5);
r.x = cvRound(m_Blob.x - 0.5*r.width);
r.y = cvRound(m_Blob.y - 0.5*r.height);
if(r.x < 0) break;
if(r.y < 0) break;
if(r.x+r.width >= pImgFG->width) break;
if(r.y+r.height >= pImgFG->height) break;
if(r.height < 5 || r.width < 5) break;
cvMoments( cvGetSubRect(pImgFG,&mat,r), &m, 0 );
M00 = cvGetSpatialMoment( &m, 0, 0 );
if(M00 <= 0 ) break;
X = cvGetSpatialMoment( &m, 1, 0 )/M00;
Y = cvGetSpatialMoment( &m, 0, 1 )/M00;
XX = (cvGetSpatialMoment( &m, 2, 0 )/M00) - X*X;
YY = (cvGetSpatialMoment( &m, 0, 2 )/M00) - Y*Y;
NewBlob = cvBlob(r.x+(float)X,r.y+(float)Y,(float)(4*sqrt(XX)),(float)(4*sqrt(YY)));
NewBlob.w = Alpha*NewBlob.w+m_Blob.w*(1-Alpha);
NewBlob.h = Alpha*NewBlob.h+m_Blob.h*(1-Alpha);
m_Blob.w = MAX(NewBlob.w,5);
m_Blob.h = MAX(NewBlob.h,5);
break;
} /* Update size if no collision. */
return &m_Blob;
}; /* CvBlobTrackerOneMSFG::Process */
virtual double GetConfidence(CvBlob* pBlob, IplImage* pImg, IplImage* /*pImgFG*/ = NULL, IplImage* pImgUnusedReg = NULL)
{
double S = 0.2;
double B = GetBhattacharyya(pImg, pImgUnusedReg, pBlob, &m_HistTemp);
return exp((B-1)/(2*S));
}; /*CvBlobTrackerOneMSFG::*/
virtual void Update(CvBlob* pBlob, IplImage* pImg, IplImage* pImgFG = NULL)
{ /* Update histogram: */
UpdateModelHist(pImg, pImgFG, pBlob?pBlob:&m_Blob);
} /*CvBlobTrackerOneMSFG::*/
virtual void Release(){delete this;};
virtual void SetCollision(int CollisionFlag)
{
m_Collision = CollisionFlag;
}
virtual void SaveState(CvFileStorage* fs)
{
cvWriteStruct(fs, "Blob", &m_Blob, "ffffi");
cvWriteInt(fs,"Collision", m_Collision);
cvWriteInt(fs,"HistVolume", cvRound(m_HistModel.m_HistVolume));
cvWrite(fs,"Hist", m_HistModel.m_pHist);
};
virtual void LoadState(CvFileStorage* fs, CvFileNode* node)
{
CvMat* pM;
cvReadStructByName(fs, node, "Blob",&m_Blob, "ffffi");
m_Collision = cvReadIntByName(fs,node,"Collision",m_Collision);
pM = (CvMat*)cvRead(fs,cvGetFileNodeByName(fs,node,"Hist"));
if(pM)
{
m_HistModel.m_pHist = pM;
m_HistModel.m_HistVolume = (float)cvSum(pM).val[0];
}
};
}; /*CvBlobTrackerOneMSFG*/
#if 0
void CvBlobTrackerOneMSFG::CollectHist(IplImage* pImg, IplImage* pMask, CvBlob* pBlob, DefHist* pHist)
{
int UsePrecalculatedKernel = 0;
int BW = cvRound(pBlob->w);
int BH = cvRound(pBlob->h);
DefHistType Volume = 0;
int x0 = cvRound(pBlob->x - BW*0.5);
int y0 = cvRound(pBlob->y - BH*0.5);
int x,y;
UsePrecalculatedKernel = (BW == m_ObjSize.width && BH == m_ObjSize.height ) ;
//cvZero(pHist);
cvSet(pHist->m_pHist,cvScalar(1.0/m_BinNumTotal)); /* no zero bins, all bins have very small value*/
Volume = 1;
assert(BW < pImg->width);
assert(BH < pImg->height);
if((x0+BW)>=pImg->width) BW=pImg->width-x0-1;
if((y0+BH)>=pImg->height) BH=pImg->height-y0-1;
if(x0<0){ x0=0;}
if(y0<0){ y0=0;}
if(m_Dim == 3)
{
for(y=0; y<BH; ++y)
{
unsigned char* pImgData = &CV_IMAGE_ELEM(pImg,unsigned char,y+y0,x0*3);
unsigned char* pMaskData = pMask?(&CV_IMAGE_ELEM(pMask,unsigned char,y+y0,x0)):NULL;
DefHistType* pKernelData = NULL;
if(UsePrecalculatedKernel)
{
pKernelData = ((DefHistType*)CV_MAT_ELEM_PTR_FAST(m_KernelHist[0],y,0,sizeof(DefHistType)));
}
for(x=0; x<BW; ++x, pImgData+=3)
{
DefHistType K;
int index = HIST_INDEX(pImgData);
assert(index >= 0 && index < pHist->m_pHist->cols);
if(UsePrecalculatedKernel)
{
K = pKernelData[x];
}
else
{
float dx = (x+x0-pBlob->x)/(pBlob->w*0.5);
float dy = (y+y0-pBlob->y)/(pBlob->h*0.5);
double r2 = dx*dx+dy*dy;
K = GetKernelHist(r2);
}
if(pMaskData)
{
K *= pMaskData[x]*0.003921568627450980392156862745098;
}
Volume += K;
((DefHistType*)(pHist->m_pHist->data.ptr))[index] += K;
} /* Next column. */
} /* Next row. */
} /* if m_Dim == 3. */
pHist->m_HistVolume = Volume;
}; /* CollectHist */
#endif
static CvBlobTrackerOne* cvCreateBlobTrackerOneMSFG()
{
return (CvBlobTrackerOne*) new CvBlobTrackerOneMSFG;
}
CvBlobTracker* cvCreateBlobTrackerMSFG()
{
return cvCreateBlobTrackerList(cvCreateBlobTrackerOneMSFG);
}
/* Create specific tracker without FG
* usin - just simple mean-shift tracker: */
class CvBlobTrackerOneMS:public CvBlobTrackerOneMSFG
{
public:
CvBlobTrackerOneMS()
{
SetParam("FGWeight",0);
DelParam("FGWeight");
SetModuleName("MS");
};
};
static CvBlobTrackerOne* cvCreateBlobTrackerOneMS()
{
return (CvBlobTrackerOne*) new CvBlobTrackerOneMS;
}
CvBlobTracker* cvCreateBlobTrackerMS()
{
return cvCreateBlobTrackerList(cvCreateBlobTrackerOneMS);
}
typedef struct DefParticle
{
CvBlob blob;
float Vx,Vy;
double W;
} DefParticle;
class CvBlobTrackerOneMSPF:public CvBlobTrackerOneMS
{
private:
/* parameters */
int m_ParticleNum;
float m_UseVel;
float m_SizeVar;
float m_PosVar;
CvSize m_ImgSize;
CvBlob m_Blob;
DefParticle* m_pParticlesPredicted;
DefParticle* m_pParticlesResampled;
CvRNG m_RNG;
#ifdef _OPENMP
int m_ThreadNum;
DefHist* m_HistForParalel;
#endif
public:
virtual void SaveState(CvFileStorage* fs)
{
CvBlobTrackerOneMS::SaveState(fs);
cvWriteInt(fs,"ParticleNum",m_ParticleNum);
cvWriteStruct(fs,"ParticlesPredicted",m_pParticlesPredicted,"ffffiffd",m_ParticleNum);
cvWriteStruct(fs,"ParticlesResampled",m_pParticlesResampled,"ffffiffd",m_ParticleNum);
};
virtual void LoadState(CvFileStorage* fs, CvFileNode* node)
{
//CvMat* pM;
CvBlobTrackerOneMS::LoadState(fs,node);
m_ParticleNum = cvReadIntByName(fs,node,"ParticleNum",m_ParticleNum);
if(m_ParticleNum>0)
{
Realloc();
printf("sizeof(DefParticle) is %d\n", (int)sizeof(DefParticle));
cvReadStructByName(fs,node,"ParticlesPredicted",m_pParticlesPredicted,"ffffiffd");
cvReadStructByName(fs,node,"ParticlesResampled",m_pParticlesResampled,"ffffiffd");
}
};
CvBlobTrackerOneMSPF()
{
m_pParticlesPredicted = NULL;
m_pParticlesResampled = NULL;
m_ParticleNum = 200;
AddParam("ParticleNum",&m_ParticleNum);
CommentParam("ParticleNum","Number of particles");
Realloc();
m_UseVel = 0;
AddParam("UseVel",&m_UseVel);
CommentParam("UseVel","Percent of particles which use velocity feature");
m_SizeVar = 0.05f;
AddParam("SizeVar",&m_SizeVar);
CommentParam("SizeVar","Size variation (in object size)");
m_PosVar = 0.2f;
AddParam("PosVar",&m_PosVar);
CommentParam("PosVar","Position variation (in object size)");
m_RNG = cvRNG(0);
SetModuleName("MSPF");
#ifdef _OPENMP
{
m_ThreadNum = omp_get_num_procs();
m_HistForParalel = new DefHist[m_ThreadNum];
}
#endif
}
~CvBlobTrackerOneMSPF()
{
if(m_pParticlesResampled)cvFree(&m_pParticlesResampled);
if(m_pParticlesPredicted)cvFree(&m_pParticlesPredicted);
#ifdef _OPENMP
if(m_HistForParalel) delete[] m_HistForParalel;
#endif
}
private:
void Realloc()
{
if(m_pParticlesResampled)cvFree(&m_pParticlesResampled);
if(m_pParticlesPredicted)cvFree(&m_pParticlesPredicted);
m_pParticlesPredicted = (DefParticle*)cvAlloc(sizeof(DefParticle)*m_ParticleNum);
m_pParticlesResampled = (DefParticle*)cvAlloc(sizeof(DefParticle)*m_ParticleNum);
}; /* Realloc*/
void DrawDebug(IplImage* pImg, IplImage* /*pImgFG*/)
{
int k;
for(k=0; k<2; ++k)
{
DefParticle* pBP = k?m_pParticlesResampled:m_pParticlesPredicted;
//const char* name = k?"MSPF resampled particle":"MSPF Predicted particle";
IplImage* pI = cvCloneImage(pImg);
int h,hN = m_ParticleNum;
CvBlob C = cvBlob(0,0,0,0);
double WS = 0;
for(h=0; h<hN; ++h)
{
CvBlob B = pBP[h].blob;
int CW = cvRound(255*pBP[h].W);
CvBlob* pB = &B;
int x = cvRound(CV_BLOB_RX(pB)), y = cvRound(CV_BLOB_RY(pB));
CvSize s = cvSize(MAX(1,x), MAX(1,y));
double W = pBP[h].W;
C.x += pB->x;
C.y += pB->y;
C.w += pB->w;
C.h += pB->h;
WS+=W;
s = cvSize(1,1);
cvEllipse( pI,
cvPointFrom32f(CV_BLOB_CENTER(pB)),
s,
0, 0, 360,
CV_RGB(CW,0,0), 1 );
} /* Next hypothesis. */
C.x /= hN;
C.y /= hN;
C.w /= hN;
C.h /= hN;
cvEllipse( pI,
cvPointFrom32f(CV_BLOB_CENTER(&C)),
cvSize(cvRound(C.w*0.5),cvRound(C.h*0.5)),
0, 0, 360,
CV_RGB(0,0,255), 1 );
cvEllipse( pI,
cvPointFrom32f(CV_BLOB_CENTER(&m_Blob)),
cvSize(cvRound(m_Blob.w*0.5),cvRound(m_Blob.h*0.5)),
0, 0, 360,
CV_RGB(0,255,0), 1 );
//cvNamedWindow(name,0);
//cvShowImage(name,pI);
cvReleaseImage(&pI);
} /* */
//printf("Blob %d, point (%.1f,%.1f) size (%.1f,%.1f)\n",m_Blob.ID,m_Blob.x,m_Blob.y,m_Blob.w,m_Blob.h);
} /* ::DrawDebug */
private:
void Prediction()
{
int p;
for(p=0; p<m_ParticleNum; ++p)
{ /* "Prediction" of particle: */
//double t;
float r[5];
CvMat rm = cvMat(1,5,CV_32F,r);
cvRandArr(&m_RNG,&rm,CV_RAND_NORMAL,cvScalar(0),cvScalar(1));
m_pParticlesPredicted[p] = m_pParticlesResampled[p];
if(cvRandReal(&m_RNG)<0.5)
{ /* Half of particles will predict based on external blob: */
m_pParticlesPredicted[p].blob = m_Blob;
}
if(cvRandReal(&m_RNG)<m_UseVel)
{ /* Predict moving particle by usual way by using speed: */
m_pParticlesPredicted[p].blob.x += m_pParticlesPredicted[p].Vx;
m_pParticlesPredicted[p].blob.y += m_pParticlesPredicted[p].Vy;
}
else
{ /* Stop several particles: */
m_pParticlesPredicted[p].Vx = 0;
m_pParticlesPredicted[p].Vy = 0;
}
{ /* Update position: */
float S = (m_Blob.w + m_Blob.h)*0.5f;
m_pParticlesPredicted[p].blob.x += m_PosVar*S*r[0];
m_pParticlesPredicted[p].blob.y += m_PosVar*S*r[1];
/* Update velocity: */
m_pParticlesPredicted[p].Vx += (float)(m_PosVar*S*0.1*r[3]);
m_pParticlesPredicted[p].Vy += (float)(m_PosVar*S*0.1*r[4]);
}
/* Update size: */
m_pParticlesPredicted[p].blob.w *= (1+m_SizeVar*r[2]);
m_pParticlesPredicted[p].blob.h *= (1+m_SizeVar*r[2]);
/* Truncate size of particle: */
if(m_pParticlesPredicted[p].blob.w > m_ImgSize.width*0.5f)
{
m_pParticlesPredicted[p].blob.w = m_ImgSize.width*0.5f;
}
if(m_pParticlesPredicted[p].blob.h > m_ImgSize.height*0.5f)
{
m_pParticlesPredicted[p].blob.h = m_ImgSize.height*0.5f;
}
if(m_pParticlesPredicted[p].blob.w < 1 )
{
m_pParticlesPredicted[p].blob.w = 1;
}
if(m_pParticlesPredicted[p].blob.h < 1)
{
m_pParticlesPredicted[p].blob.h = 1;
}
} /* "Prediction" of particle. */
} /* Prediction */
void UpdateWeightsMS(IplImage* pImg, IplImage* /*pImgFG*/)
{
int p;
#ifdef _OPENMP
if( m_HistForParalel[0].m_pHist==NULL || m_HistForParalel[0].m_pHist->cols != m_BinNumTotal)
{
int t;
for(t=0; t<m_ThreadNum; ++t)
m_HistForParalel[t].Resize(m_BinNumTotal);
}
#endif
#ifdef _OPENMP
#pragma omp parallel for num_threads(m_ThreadNum) schedule(runtime)
#endif
for(p=0;p<m_ParticleNum;++p)
{ /* Calculate weights for particles: */
double S = 0.2;
double B = 0;
#ifdef _OPENMP
assert(omp_get_thread_num()<m_ThreadNum);
#endif
B = GetBhattacharyya(
pImg, NULL,
&(m_pParticlesPredicted[p].blob)
#ifdef _OPENMP
,&(m_HistForParalel[omp_get_thread_num()])
#endif
);
m_pParticlesPredicted[p].W *= exp((B-1)/(2*S));
} /* Calculate weights for particles. */
}
void UpdateWeightsCC(IplImage* /*pImg*/, IplImage* /*pImgFG*/)
{
int p;
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(p=0; p<m_ParticleNum; ++p)
{ /* Calculate weights for particles: */
double W = 1;
m_pParticlesPredicted[p].W *= W;
} /* Calculate weights for particles. */
}
void Resample()
{ /* Resample particle: */
int p;
double Sum = 0;
for(p=0; p<m_ParticleNum; ++p)
{
Sum += m_pParticlesPredicted[p].W;
}
for(p=0; p<m_ParticleNum; ++p)
{
double T = Sum * cvRandReal(&m_RNG); /* Set current random threshold for cululative weight. */
int p2;
double Sum2 = 0;
for(p2=0; p2<m_ParticleNum; ++p2)
{
Sum2 += m_pParticlesPredicted[p2].W;
if(Sum2 >= T)break;
}
if(p2>=m_ParticleNum)p2=m_ParticleNum-1;
m_pParticlesResampled[p] = m_pParticlesPredicted[p2];
m_pParticlesResampled[p].W = 1;
} /* Find next particle. */
} /* Resample particle. */
public:
virtual void Init(CvBlob* pBlobInit, IplImage* pImg, IplImage* pImgFG = NULL)
{
int i;
CvBlobTrackerOneMSFG::Init(pBlobInit, pImg, pImgFG);
DefParticle PP;
PP.W = 1;
PP.Vx = 0;
PP.Vy = 0;
PP.blob = pBlobInit[0];
for(i=0;i<m_ParticleNum;++i)
{
m_pParticlesPredicted[i] = PP;
m_pParticlesResampled[i] = PP;
}
m_Blob = pBlobInit[0];
} /* CvBlobTrackerOneMSPF::Init*/
virtual CvBlob* Process(CvBlob* pBlobPrev, IplImage* pImg, IplImage* pImgFG = NULL)
{
int p;
m_ImgSize.width = pImg->width;
m_ImgSize.height = pImg->height;
m_Blob = pBlobPrev[0];
{ /* Check blob size and realloc kernels if it is necessary: */
int w = cvRound(m_Blob.w);
int h = cvRound(m_Blob.h);
if( w != m_ObjSize.width || h!=m_ObjSize.height)
{
ReAllocKernel(w,h);
/* After this ( w != m_ObjSize.width || h!=m_ObjSize.height) should be false. */
}
} /* Check blob size and realloc kernels if it is necessary. */
Prediction();
#ifdef REPORT_TICKS
int64 ticks = cvGetTickCount();
#endif
UpdateWeightsMS(pImg, pImgFG);
#ifdef REPORT_TICKS
ticks = cvGetTickCount() - ticks;
fprintf(stderr, "PF UpdateWeights, %d ticks\n", (int)ticks);
ticks = cvGetTickCount();
#endif
Resample();
#ifdef REPORT_TICKS
ticks = cvGetTickCount() - ticks;
fprintf(stderr, "PF Resampling, %d ticks\n", (int)ticks);
#endif
{ /* Find average result: */
float x = 0;
float y = 0;
float w = 0;
float h = 0;
float Sum = 0;
DefParticle* pP = m_pParticlesResampled;
for(p=0; p<m_ParticleNum; ++p)
{
float W = (float)pP[p].W;
x += W*pP[p].blob.x;
y += W*pP[p].blob.y;
w += W*pP[p].blob.w;
h += W*pP[p].blob.h;
Sum += W;
}
if(Sum>0)
{
m_Blob.x = x / Sum;
m_Blob.y = y / Sum;
m_Blob.w = w / Sum;
m_Blob.h = h / Sum;
}
} /* Find average result. */
if(m_Wnd)
{
DrawDebug(pImg, pImgFG);
}
return &m_Blob;
} /* CvBlobTrackerOneMSPF::Process */
virtual void SkipProcess(CvBlob* pBlob, IplImage* /*pImg*/, IplImage* /*pImgFG*/ = NULL)
{
int p;
for(p=0; p<m_ParticleNum; ++p)
{
m_pParticlesResampled[p].blob = pBlob[0];
m_pParticlesResampled[p].Vx = 0;
m_pParticlesResampled[p].Vy = 0;
m_pParticlesResampled[p].W = 1;
}
}
virtual void Release(){delete this;};
virtual void ParamUpdate()
{
Realloc();
}
}; /* CvBlobTrackerOneMSPF */
CvBlobTrackerOne* cvCreateBlobTrackerOneMSPF();
CvBlobTrackerOne* cvCreateBlobTrackerOneMSPF()
{
return (CvBlobTrackerOne*) new CvBlobTrackerOneMSPF;
}
CvBlobTracker* cvCreateBlobTrackerMSPF()
{
return cvCreateBlobTrackerList(cvCreateBlobTrackerOneMSPF);
}
Reference in New Issue View Git Blame Copy Permalink