void TLD::bbPredict(const vector<cv::Point2f>& points1,const vector<cv::Point2f>& points2,
const BoundingBox& bb1,BoundingBox& bb2) {
int npoints = (int)points1.size();
vector<float> xoff(npoints);
vector<float> yoff(npoints);
printf("tracked points : %d\n",npoints);
for (int i=0;i<npoints;i++){
xoff[i]=points2[i].x-points1[i].x;
yoff[i]=points2[i].y-points1[i].y;
}
float dx = median(xoff);
float dy = median(yoff);
float s;
if (npoints>1){
vector<float> d;
d.reserve(npoints*(npoints-1)/2);
for (int i=0;i<npoints;i++){
for (int j=i+1;j<npoints;j++){
d.push_back(norm(points2[i]-points2[j])/norm(points1[i]-points1[j]));
}
}
s = median(d);
}
else {
s = 1.0;
}
float s1 = 0.5*(s-1)*bb1.width;
float s2 = 0.5*(s-1)*bb1.height;
printf("s= %f s1= %f s2= %f \n",s,s1,s2);
bb2.x = round( bb1.x + dx -s1);
bb2.y = round( bb1.y + dy -s2);
bb2.width = round(bb1.width*s);
bb2.height = round(bb1.height*s);
printf("predicted bb: %d %d %d %d\n",bb2.x,bb2.y,bb2.br().x,bb2.br().y);
}
void Detector::generateScanGrids(const BoundingBox& initBoundingBox)
{
vector<Size> gridSizes;
Size bbSize = initBoundingBox.size();
double scale = 1.0;
while (bbSize.width <= frame.cols && bbSize.height <= frame.rows)
{
gridSizes.push_back(bbSize);
scale *= 1.2;
bbSize = Size((int)(initBoundingBox.width * scale + 0.5), (int)(initBoundingBox.height * scale + 0.5));
}
scale = 1.0 / STEP_S;
bbSize = Size((int)(initBoundingBox.width * scale + 0.5), (int)(initBoundingBox.height * scale + 0.5));
//while (bbSize.width * bbSize.height >= MIN_BB_AREA)
while (bbSize.width >= MIN_BB_SIDE && bbSize.height >= MIN_BB_SIDE)
{
gridSizes.push_back(bbSize);
scale /= 1.2;
bbSize = Size((int)(initBoundingBox.width * scale + 0.5), (int)(initBoundingBox.height * scale + 0.5));
}
//int stepH = (int)(STEP_H * frame.cols + 0.5);
//int stepV = (int)(STEP_V * frame.rows + 0.5);
BoundingBox patch;
for (vector<Size>::const_iterator scaleIt = gridSizes.begin(); scaleIt != gridSizes.end(); ++scaleIt)
{
patch.width = scaleIt->width;
patch.height = scaleIt->height;
int minSide = min(scaleIt->width, scaleIt->height);
//int stepH = (int)(STEP_H * scaleIt->width + 0.5);
//int stepV = (int)(STEP_V * scaleIt->height + 0.5);
int stepH = (int)(STEP_H * minSide + 0.5);
int stepV = (int)(STEP_V * minSide + 0.5);
#ifdef DEBUG
assert(stepH >= 1 && stepV >= 1);
#endif
for (patch.y = 0; patch.br().y < frame.rows; patch.y += stepV)
{
for (patch.x = 0; patch.br().x < frame.cols; patch.x += stepH)
{
patch.refreshOverlap(initBoundingBox);
scanGrids.push_back(patch);
}
}
}
}
//利用剩下的不到一半的跟踪点输入来预测bounding box在当前帧的位置和大小
void TLD::bbPredict(const vector<cv::Point2f>& points1,const vector<cv::Point2f>& points2,
const BoundingBox& bb1,BoundingBox& bb2) {
int npoints = (int)points1.size();
vector<float> xoff(npoints);//位移
vector<float> yoff(npoints);
printf("tracked points : %d\n",npoints);
for (int i=0;i<npoints;i++){//计算每个特征点在两帧之间的位移
xoff[i]=points2[i].x-points1[i].x;
yoff[i]=points2[i].y-points1[i].y;
}
float dx = median(xoff);//计算位移的中值
float dy = median(yoff);
float s;
//计算bounding box尺度scale的变化
//通过计算当前特征点相互间的距离与上一帧特征点相互间的距离的
//比值，以比值的中值作为尺度的变化因子
if (npoints>1){
vector<float> d;
d.reserve(npoints*(npoints-1)/2);
for (int i=0;i<npoints;i++){
for (int j=i+1;j<npoints;j++){
d.push_back(norm(points2[i]-points2[j])/norm(points1[i]-points1[j]));
}
}
s = median(d);
}
else {
s = 1.0;
}
float s1 = 0.5*(s-1)*bb1.width;
float s2 = 0.5*(s-1)*bb1.height;
printf("s= %f s1= %f s2= %f \n",s,s1,s2);
//得到当前bounding box的位置与大小信息
//当前box的x坐标 = 前一帧box的x坐标 + 全部特征点位移的中值 - 当前box宽的一半
bb2.x = round( bb1.x + dx -s1);
bb2.y = round( bb1.y + dy -s2);
bb2.width = round(bb1.width*s);
bb2.height = round(bb1.height*s);
printf("predicted bb: %d %d %d %d\n",bb2.x,bb2.y,bb2.br().x,bb2.br().y);
}
double PatchVariance::computeVariance(const BoundingBox& patch) const
{
double sum = integralImg.at<double>(patch.br())
- integralImg.at<double>(patch.br().y, patch.x)
- integralImg.at<double>(patch.y, patch.br().x)
+ integralImg.at<double>(patch.tl());
double sumSq = sqIntegralImg.at<double>(patch.br())
- sqIntegralImg.at<double>(patch.br().y, patch.x)
- sqIntegralImg.at<double>(patch.y, patch.br().x)
+ sqIntegralImg.at<double>(patch.tl());
return sumSq / patch.area() - pow(sum / patch.area(), 2);
}
