/**
* @brief Is Camshift Tracker updated or not?
* @param img
* @param obj
* @param hist
* @return bool, whether Camshift tracker has been updated or not
*/
bool CTrackingAlgs::CamshiftUpdateTracker( const cv::Mat& img,
const cv::Rect& obj,
cv::MatND& hist)
{
cv::Mat hsv, hue, mask, backproject;
cv::cvtColor( img, hsv, CV_BGR2HSV );
int _vmin = CTrackingAlgs::vmin, _vmax = CTrackingAlgs::vmax;
cv::inRange( hsv, cv::Scalar(0,CTrackingAlgs::smin,MIN(_vmin,_vmax),0),
cv::Scalar(180,256,MAX(_vmin,_vmax),0), mask );
std::vector<cv::Mat> vhsv(3);
cv::split( hsv, vhsv );
vhsv[0].copyTo(hue);
double max_val = 0.f;
cv::Mat roi = hue( cv::Range(obj.y, obj.y+obj.height),
cv::Range(obj.x, obj.x+obj.width) );
cv::Mat roi_mask= mask( cv::Range(obj.y, obj.y+obj.height),
cv::Range(obj.x, obj.x+obj.width) );
cv::calcHist( &roi, 1, CTrackingAlgs::channels, roi_mask,
hist, 1, CTrackingAlgs::histSize, CTrackingAlgs::ranges,
true, // the histogram is uniform
false );
cv::minMaxLoc(hist, 0, &max_val, 0, 0);
hist.convertTo(hist, hist.type(), (max_val ? 255. / max_val : 0.), 0);
return true;
}
/**
* Convert cv::MatND to MxArray
* @param mat cv::MatND object
* @param classid classid of mxArray. e.g., mxDOUBLE_CLASS. When mxUNKNOWN_CLASS
* is specified, classid will be automatically determined from
* the type of cv::Mat. default: mxUNKNOWN_CLASS
* @return MxArray object
*/
MxArray::MxArray(const cv::MatND& mat, mxClassID classid)
{
if (mat.datastart==mat.dataend) {
p_ = mxCreateNumericArray(0,0,mxDOUBLE_CLASS,mxREAL);
return;
}
// Create a new mxArray
int nchannels = mat.channels();
const int* dims_ = mat.size;
vector<mwSize> d(dims_,dims_+mat.dims);
d.push_back(nchannels);
classid = (classid == mxUNKNOWN_CLASS) ? ClassIDOf[mat.depth()] : classid;
std::swap(d[0],d[1]);
p_ = mxCreateNumericArray(d.size(),&d[0],classid,mxREAL);
if (!p_)
mexErrMsgIdAndTxt("mexopencv:error","Allocation error");
// Copy
int depth = CV_MAKETYPE(DepthOf[classid],1);
cv::MatND m(mat.dims,mat.size,CV_MAKETYPE(depth,1));
uchar* _data = m.data;
uchar* _datastart = m.datastart;
uchar* _dataend = m.dataend;
m.data = m.datastart = reinterpret_cast<uchar*>(mxGetData(p_));
m.dataend = reinterpret_cast<uchar*>(
reinterpret_cast<size_t>(mxGetData(p_))+mxGetElementSize(p_)*numel());
mat.convertTo(m,CV_MAKETYPE(depth,1));
m.data = _data;
m.datastart = _datastart;
m.dataend = _dataend;
}
std::vector<jsk_recognition_msgs::HistogramWithRangeBin>
cvMatNDToHistogramWithRangeBinArray(const cv::MatND& cv_hist, float min_value, float max_value)
{
std::vector<jsk_recognition_msgs::HistogramWithRangeBin> bins(cv_hist.total());
const float bin_width = (max_value - min_value) / cv_hist.total();
for (size_t i = 0; i < cv_hist.total(); i++) {
const float left = i * bin_width + min_value;
const float right = (i + 1) * bin_width + min_value;
jsk_recognition_msgs::HistogramWithRangeBin bin;
bin.min_value = left;
bin.max_value = right;
bin.count = cv_hist.at<float>(0, i);
bins[i] = bin;
}
return bins;
}
cv::MatND MxArray::toMatND(int depth, bool transpose) const
{
// Create cv::MatND object (of the specified depth), equivalent to mxArray
std::vector<int> d(dims(), dims()+ndims());
std::swap(d[0], d[1]);
depth = (depth == CV_USRTYPE1) ? DepthOf[classID()] : depth;
cv::MatND mat(d.size(), &d[0], CV_MAKETYPE(depth, 1));
// Copy from mxArray to cv::MatND (converting to specified depth)
const int type = CV_MAKETYPE(DepthOf[classID()], 1); // source type
const cv::MatND m(d.size(), &d[0], type, mxGetData(p_)); // only Mat header
// Read from mxArray through m, writing into mat
m.convertTo(mat, CV_MAKETYPE(depth, 1));
// transpose cv::MatND if needed
if (mat.dims==2 && transpose)
mat = mat.t();
return mat;
}
// We don't know if MatND uses the little-endian dimension order (i.e. dim=0: lowest dimension, dim=nd-1: highest dimension) or the big-endian order. Therefore we will detect the order from the instance.
// Note: http://code.google.com/p/pyopencv/issues/detail?id=18
bool get_array_data_arrangement(cv::MatND const &inst, sdcpp::array_data_arrangement &result)
{
if(!inst.flags) return false;
bool endianness = (inst.dims >= 2) && (inst.step[0] > inst.step[1]);
bool multichannel = inst.channels() > 1;
int i;
result.item_size = inst.elemSize1();
result.ndim = inst.dims + multichannel;
result.size.resize(result.ndim);
result.stride.resize(result.ndim);
if(multichannel)
{
result.size[inst.dims] = inst.channels();
result.stride[inst.dims] = inst.elemSize1();
}
if(endianness)
{
result.total_size = inst.size[0]*inst.step[0];
for(i = 0; i < inst.dims; ++i)
{
result.size[i] = inst.size[i];
result.stride[i] = inst.step[i];
}
}
else
{
result.total_size = inst.size[inst.dims-1]*inst.step[inst.dims-1];
for(i = 0; i < inst.dims; ++i)
{
result.size[i] = inst.size[inst.dims-1-i];
result.stride[i] = inst.step[inst.dims-1-i];
}
}
return true;
}
double HullToObjectModule::histogramDistance(cv::MatND h1, cv::MatND h2) {
cv::MatND d(h1.size(), CV_32FC1);
cv::absdiff(h1, h2, d);
cv::Scalar s = cv::sum(d);
return s.val[0];
}
double setDegu::histogramDistance(cv::MatND h1, cv::MatND h2) {
cv::MatND d(h1.size(), CV_32FC1);
cv::absdiff(h1, h2, d);
cv::Scalar s = cv::sum(d);
return s.val[0];
}
double regionalSegmentationModule::histogramDistance(cv::MatND h1, cv::MatND h2) {
cv::MatND d(h1.size(), CV_32FC1);
cv::absdiff(h1, h2, d);
cv::Scalar s = cv::sum(d);
return s.val[0];
}
