vector<float> HOGDetectorGPU::getHOGdescriptors(cv::HOGDescriptor& hog, Mat& image){
resize(image, image, Size(WIN_X,WIN_Y) );
Mat img;
cvtColor(image, img, CV_RGB2GRAY);
hog.winSize.height = WIN_X;
hog.winSize.width = WIN_Y;
// Size(128,64), //winSize
// Size(16,16), //blocksize
// Size(8,8), //blockStride,
// Size(8,8), //cellSize,
// 9, //nbins,
// 0, //derivAper,
// -1, //winSigma,
// 0, //histogramNormType,
// 0.2, //L2HysThresh,
// 0 //gammal correction,
// //nlevels=64
//);
vector<float> descriptorsValues;
vector<Point> locations;
hog.compute( img, descriptorsValues, Size(0,0), Size(0,0), locations);
cout << "HOG descriptor size is " << hog.getDescriptorSize() << endl;
cout << "img dimensions: " << img.cols << " width x " << img.rows << "height" << endl;
cout << "Found " << descriptorsValues.size() << " descriptor values" << endl;
cout << "Nr of locations specified : " << locations.size() << endl;
return descriptorsValues;
}
people_detection_hog(const options& in_opt):
opt(in_opt.person_d)
{
scenter_x=in_opt.width*opt.scale/2.0;
scenter_y=in_opt.height*opt.scale/2.0;
center_x=in_opt.width/2.0;
center_y=in_opt.height/2.0;
hog.setSVMDetector(hog.getDefaultPeopleDetector());
}
static void calculateFeaturesFromInput( cv::Mat imageData, std::vector< float >& featureVector, cv::HOGDescriptor& hog )
{
std::vector< cv::Point > locations;
hog.compute( imageData, featureVector, winStride, trainingPadding, locations );
imageData.release();
}
int main(int argc, char *argv[]) {
hog.setSVMDetector(cv::HOGDescriptor::getDefaultPeopleDetector());
//hog.setSVMDetector(cv::HOGDescriptor::getDaimlerPeopleDetector());
cv::Mat input_img;
if (argc == 2) {
//printf("Usage %s [file]\n", argv[0]);
// return -1;
std::string file(argv[1]);
input_img = cv::imread(file, CV_LOAD_IMAGE_COLOR);
process_image(input_img);
cv::waitKey(0);
}
//cv::VideoCapture cap(0);
cv::VideoCapture cap;
cap.open("/home/arnaud/Desktop/Baudet-Rob.mp4");
if (!cap.isOpened()) {
printf("Cannot open camera!"); return -1;
}
while(true) {
cap >> input_img;
cv::resize(input_img, input_img, cv::Size(320, 240));
process_image(input_img);
cv::waitKey(15);
}
return 0;
}
std::vector<cv::Rect> hogDetection(cv::Mat sequence, cv::HOGDescriptor hog)
{
std::vector<cv::Rect> found;
hog.detectMultiScale(sequence, found, 0, cv::Size(8,8), cv::Size(32,32), 1.05, 2);
return found;
}
int process_image(cv::Mat & input_img) {
std::vector<cv::Rect> found_rectangles;
hog.detectMultiScale(input_img, found_rectangles);
for (unsigned int rec_idx = 0; rec_idx < found_rectangles.size(); ++rec_idx)
cv::rectangle(input_img, found_rectangles[rec_idx], CV_RGB(255,0,0), 2);
cv::imwrite("input_img.png", input_img);
cv::imshow("input_img", input_img);
} // end process_image()
void peopleDetectApp::setup()
{
hog.setSVMDetector(cv::HOGDescriptor::getDefaultPeopleDetector());
capture = Capture(640, 480);
capture.start();
stillSurface = loadImage(loadAsset( "people_in_park.jpg" ) );
stillTex = gl::Texture(stillSurface);
}
void ocvOpticalFlowApp::setup()
{
mDrawPoints = true;
// mCapture = Capture( 640, 480 );
// mCapture.start();
fs::path moviePath = getOpenFilePath();
if( ! moviePath.empty() )
loadMovieFile( moviePath );
hog.setSVMDetector( cv::HOGDescriptor::getDefaultPeopleDetector() );
}
cv::Mat computeHOG(const cv::Mat &img_gray, cv::HOGDescriptor &d) {
int correct_width = img_gray.cols - (img_gray.cols - d.blockSize.width) % d.blockStride.width;
int correct_height = img_gray.rows - (img_gray.rows - d.blockSize.height) % d.blockStride.height;
d.winSize = cv::Size(correct_width, correct_height);
//Scale input image to the correct size
cv::Mat resize_img;
cv::resize(img_gray, resize_img, d.winSize);
std::vector<float> descriptorsValues;
std::vector<cv::Point> locations;
d.compute(resize_img, descriptorsValues, cv::Size(0,0), cv::Size(0,0), locations);
cv::Mat matDescriptorsValues(descriptorsValues, true);
return matDescriptorsValues;
}
void peopleDetectApp::update()
{
if( getElapsedFrames() % 30 == 0 )
{
if(stillSurface) {
//cv::Mat newFrame = toOcv( Channel( capture.getSurface() )); // capture the frame
cv::Mat newFrame = toOcv( Channel(stillSurface));
vector<cv::Rect> found, found_filtered;
// run the detector with default parameters. to get a higher hit-rate
// (and more false alarms, respectively), decrease the hitThreshold and
// groupThreshold (set groupThreshold to 0 to turn off the grouping completely).
//hog.detectMultiScale(newFrame, foundLocations, 0, cv::Size(8,8), cv::Size(32,32), 1.05, 2);
hog.detectMultiScale(newFrame, foundLocations, 0, cv::Size(8,8), cv::Size(32,32), 1.05, 2);
int i,j;
for( i = 0; i < foundLocations.size(); i++ )
{
cv::Rect r = foundLocations[i];
for( j = 0; j < foundLocations.size(); j++ ) {
if( j != i && (r & foundLocations[j]) == r)
break;
if( j == foundLocations.size() )
found_filtered.push_back(r);
}
}
console() << " found # " << foundLocations.size() << " filtered to " << found_filtered.size() << endl;
ciPeople.clear();
for( i = 0; i < foundLocations.size(); i++ ) {
cv::Rect r = foundLocations[i];
// the HOG detector returns slightly larger rectangles than the real objects.
// so we slightly shrink the rectangles to get a nicer output.
r.x += cvRound(r.width*0.1);
r.width = cvRound(r.width*0.8);
r.y += cvRound(r.height*0.07);
r.height = cvRound(r.height*0.8);
ciPeople.push_back(fromOcv(r));
}
}
}
}
void detect(shared_data& in_data)
{
std::vector<cv::Rect> temp_boxes;
for (auto i: in_data.im_data.im_ROI) {
temp_resize.create(in_data.im_data.image(i).rows*opt.scale, in_data.im_data.image(i).cols*opt.scale, CV_64FC3);
cv::resize(in_data.im_data.image(i), temp_resize, temp_resize.size());
//these magic numbers are from opencv
if (temp_resize.rows > 128 && temp_resize.cols > 64) {
temp_boxes.clear();
hog.detectMultiScale(temp_resize,
temp_boxes,
0,
cv::Size(8, 8),
cv::Size(0, 0),
1.05);
for (auto j: temp_boxes) {
i.x -= center_x;
i.y -= center_y;
i.x = opt.scale*i.x+scenter_x;
i.y = opt.scale*i.y+scenter_y;
j.x += i.x;
j.y += i.y;
j.x -= scenter_x;
j.y -= scenter_y;
j.x = 1.0/opt.scale*j.x+center_x;
j.y = 1.0/opt.scale*j.y+center_y;
j.height/=opt.scale;
j.width/=opt.scale;
box_safe(j, 0, 0, in_data.im_data.image.cols-1, in_data.im_data.image.rows-1);
shared_data::bbox temp;
temp = j;
temp.type_label = TYPE_PERSON;
in_data.im_boxes.push_back(temp);
}
}
}
}
void ocvOpticalFlowApp::findPeople(cv::Mat curImg)
{
cv::Mat monoImg;
cv::cvtColor(curImg, monoImg, CV_BGR2GRAY);
hog.detectMultiScale(monoImg, mFoundPeople);
}
