CvMat* HOGProcessor::calcHOGFromImage(char* filename, CvSize window, int normalization)
{
IplImage* im = cvLoadImage(filename);
//Convert Image To 64x128
IplImage* dst = cvCreateImage(window,im->depth,im->nChannels);
cvResize(im,dst);
// Tinh integral histogram
IplImage** integrals;
integrals = calculateIntegralHOG(dst);
CvMat* imgFeatureVector;
imgFeatureVector = calcHOGWindow(integrals,
cvRect(0, 0, window.width, window.height),
normalization);
for (int k = 0; k < NumOfBins; k++)
cvReleaseImage(&integrals[k]);
return imgFeatureVector;
}
void detectHOG_impl(const cv::Mat &template_img, const cv::Mat &query_img, std::vector<cv::Rect> &detections,
const double distanceThresh=50.0, const int offsetX=10, const int offsetY=10) {
int maxWidth = query_img.cols-template_img.cols;
int maxHeight = query_img.rows-template_img.rows;
cv::Mat matching_cost_map(maxHeight, maxWidth, CV_32F, std::numeric_limits<float>::max());
cv::Mat template_img_gray, query_img_gray;
cv::cvtColor(template_img, template_img_gray, cv::COLOR_BGR2GRAY);
cv::cvtColor(query_img, query_img_gray, cv::COLOR_BGR2GRAY);
// cv::equalizeHist(template_img_gray, template_img_gray);
// cv::equalizeHist(query_img_gray, query_img_gray);
#if 0
cv::HOGDescriptor hog_descriptor;
cv::Mat template_descriptors = computeHOG(template_img_gray, hog_descriptor);
for(int i = 0; i < matching_cost_map.rows; i += offsetY) {
for(int j = 0; j < matching_cost_map.cols; j += offsetX) {
cv::Mat sub_image = query_img_gray(cv::Rect(j, i, template_img.cols, template_img.rows));
cv::Mat query_descriptors = computeHOG(sub_image, hog_descriptor);
double dist = cv::norm(template_descriptors, query_descriptors, cv::NORM_L2);
matching_cost_map.at<float>(i,j) = (float) dist;
}
}
#else
std::vector<cv::Mat> template_integralImages = calculateIntegralHOG(template_img_gray, 9);
std::vector<cv::Mat> query_integralImages = calculateIntegralHOG(query_img_gray, 9);
cv::Mat template_hogCell;
cv::Rect template_roi(0, 0, template_img_gray.cols, template_img_gray.rows);
calculateHOG_rect(template_hogCell, template_integralImages, template_roi, 9, 3, 3);
for(int i = 0; i < matching_cost_map.rows; i += offsetY) {
for(int j = 0; j < matching_cost_map.cols; j += offsetX) {
cv::Mat sub_image = query_img_gray(cv::Rect(j, i, template_img.cols, template_img.rows));
cv::Rect query_roi(j, i, template_img_gray.cols, template_img_gray.rows);
cv::Mat query_hogCell;
calculateHOG_rect(query_hogCell, query_integralImages, query_roi, 9, 3, 3);
// double dist = cv::norm(template_hogCell, query_hogCell, cv::NORM_L2);
double dist = cv::compareHist(template_hogCell, query_hogCell, cv::HISTCMP_BHATTACHARYYA);
matching_cost_map.at<float>(i,j) = (float) dist;
}
}
#endif
#if DISPLAY_COST_MAP
cv::Mat matching_cost_map_display;
#endif
double minVal, maxVal;
cv::Point minLoc, maxLoc;
cv::minMaxLoc(matching_cost_map, &minVal, &maxVal, &minLoc, &maxLoc);
std::cout << "Min cost=" << minVal << " ; Max cost=" << maxVal << std::endl;
#if DISPLAY_COST_MAP
//Convert matching map for display
matching_cost_map.convertTo(matching_cost_map_display, CV_8U, 255.0/(maxVal-minVal), -255.0*minVal/(maxVal-minVal));
#endif
do {
cv::minMaxLoc(matching_cost_map, &minVal, &maxVal, &minLoc, &maxLoc);
// std::cout << "HOG distance=" << minVal << " ; distanceThresh=" << distanceThresh << std::endl;
//"Reset the location" to find other detections
#if 1
matching_cost_map.at<float>(minLoc.y, minLoc.x) = std::numeric_limits<float>::max();
#else
//Not working
int width = minLoc.x+template_img.cols < matching_cost_map.cols ? template_img.cols : matching_cost_map.cols-minLoc.x;
int height = minLoc.y+template_img.rows < matching_cost_map.rows ? template_img.rows : matching_cost_map.rows-minLoc.y;
cv::Rect detect_roi = cv::Rect(minLoc.x, minLoc.y, width, height);
cv::Mat subregion = matching_cost_map(detect_roi);
subregion = std::numeric_limits<float>::max();
#endif
//Add detection
detections.push_back(cv::Rect(minLoc.x, minLoc.y, template_img.cols, template_img.rows));
} while(minVal < distanceThresh);
#if DISPLAY_COST_MAP
cv::imshow("matching_cost_map_display", matching_cost_map_display);
// cv::waitKey(0);
#endif
}
CvMat* HoGProcessor::train_64x128(char *prefix, char *suffix,
CvSize window, int number_samples, int start_index,
int end_index, char *savexml, int canny,
int block, int normalization){
char filename[1000] = "\0", number[8];
int prefix_length;
prefix_length = strlen(prefix);
int bins = 9;
/* A default block size of 2x2 cells is considered */
int block_width = m_block.width , block_height = m_block.height;
/* Calculation of the length of a feature vector for
an image (64x128 pixels)*/
int feature_vector_length;
feature_vector_length = (((window.width - m_cell.width * block_width)/ (m_cell.width * m_fStepOverlap)) + 1) *
(((window.height - m_cell.height * block_height)
/ (m_cell.height * m_fStepOverlap)) + 1) * (9 * m_block.height * m_block.width);
/* Matrix to store the feature vectors for
all(number_samples) the training samples */
CvMat* training = cvCreateMat(number_samples,
feature_vector_length, CV_32FC1);
CvMat row;
CvMat* img_feature_vector;
IplImage** integrals;
int i = 0, j = 0;
printf("Beginning to extract HoG features\n");
strcat(filename, prefix);
/* Loop to calculate hog features for each
image one by one */
for (i = start_index; i <= end_index; i++)
{
itoa(i,number,10);
strcat(filename, number);
strcat(filename, suffix);
IplImage* im = cvLoadImage(filename);
//Convert Image To 64x128
IplImage* dst = cvCreateImage(window,im->depth,im->nChannels);
cvResize(im,dst);
/* Calculation of the integral histogram for
fast calculation of hog features*/
integrals = calculateIntegralHOG(dst);
cvGetRow(training, &row, j);
img_feature_vector
= calculateHOG_window(integrals, cvRect(0, 0,
window.width, window.height), normalization);
cvCopy(img_feature_vector, &row);
j++;
printf("%s\n", filename);
filename[prefix_length] = '\0';
for (int k = 0; k < 9; k++)
{
cvReleaseImage(&integrals[k]);
}
}
if (savexml != NULL)
{
cvSave(savexml, training);
}
return training;
}
CvRect HoGProcessor::detectObject(CvSVM *svmModel, IplImage *input, IplImage *result, CvRect rectHead, int normalization){
int StepWidth = 10;
int StepHeight = 10;
int nWindow = 3;
int scaleWidth = 2;
int scaleHeight = 2;
ImageProcessor ip;
//loai bo truong hop toc dai
if(rectHead.height > rectHead.width)
rectHead.height = rectHead.width;
CvRect rectHuman = cvRect(rectHead.x + rectHead.width/2 - rectHead.width*scaleWidth/2,
rectHead.y - 6,
rectHead.width*scaleWidth,
rectHead.height*scaleHeight);
vector<CvRect> lstRect;
CvMat* img_feature_vector;
IplImage **newIntegrals;
for(int i = 0; i < nWindow; i++)
{
for(int j = 0; j < nWindow; j++)
{
CvRect rect;
rect.width = rectHuman.width + StepWidth*i;
rect.height = rectHuman.height + StepHeight*j;
rect.x = rectHuman.x - StepWidth*i/2;
rect.y = rectHuman.y - StepHeight*j/2;
if(rect.x < 0) rect.x = 0;
if(rect.y < 0) rect.y = 0;
if(rect.x + rect.width > input->width) rect.width = input->width - rect.x;
if(rect.y + rect.height > input->height) rect.height = input->height - rect.y;
IplImage* candidate_img = ip.getSubImageAndResize(input, rect, 48, 48);
if(candidate_img)
{
newIntegrals = calculateIntegralHOG(candidate_img);
img_feature_vector = calculateHOG_window(newIntegrals,cvRect(0,0,48,48),4);
for (int k = 0; k < 9; k++)
{
cvReleaseImage(&newIntegrals[k]);
}
cvReleaseImage(newIntegrals);
cvReleaseImage(&candidate_img);
double predict_rs = svmModel->predict(img_feature_vector, true);
if(predict_rs >= -1)
lstRect.push_back(rect);
cvReleaseMat(&img_feature_vector);
}
}
}
if(lstRect.size() > 0)
{
return MergeRect(lstRect);
}
return cvRect(0,0,-1,-1);
}
CvMat* HoGProcessor::train_large(char *prefix, char *suffix,
CvSize window, int number_images,
int horizontal_scans, int vertical_scans,
int start_index, int end_index,
char *savexml, int normalization)
{
char filename[200] = "\0", number[8];
int prefix_length;
prefix_length = strlen(prefix);
int bins = 9;
/* A default block size of 2x2 cells is considered */
int block_width = m_block.width , block_height = m_block.height;
/* Calculation of the length of a feature vector for
an image (64x128 pixels)*/
int feature_vector_length;
feature_vector_length = (((window.width -
m_cell.width * block_width) / (m_cell.width * m_fStepOverlap)) + 1) *
(((window.height - m_cell.height * block_height)
/ (m_cell.height * m_fStepOverlap)) + 1) * (9 * m_block.height * m_block.width);
/* Matrix to store the feature vectors for
all(number_samples) the training samples */
CvMat* training = cvCreateMat(number_images
* horizontal_scans * vertical_scans,
feature_vector_length, CV_32FC1);
CvMat row;
CvMat* img_feature_vector;
IplImage** integrals;
int i = 0, j = 0;
strcat(filename, prefix);
printf("Beginning to extract HoG features from negative images\n");
/* Loop to calculate hog features for each
image one by one */
for (i = start_index; i <= end_index; i++)
{
itoa(i,number,10);
strcat(filename, number);
strcat(filename, suffix);
IplImage* im = cvLoadImage(filename);
IplImage* img = im;
//Chi xet nhung anh nao co kich thuoc lon hon hoac bang window
if(im->width < window.width || im->height < window.height)
{
img = cvCreateImage(window,im->depth, im->nChannels);
cvResize(im,img);
}
integrals = calculateIntegralHOG(img);
for (int l = 0; l < vertical_scans - 1; l++)
{
for (int k = 0; k < horizontal_scans - 1; k++)
{
cvGetRow(training, &row, j);
img_feature_vector = calculateHOG_window(
integrals, cvRect(window.width * k,
window.height * l, window.width,
window.height), normalization);
cvCopy(img_feature_vector, &row);
j++;
}
cvGetRow(training, &row, j);
img_feature_vector = calculateHOG_window(
integrals, cvRect(img->width - window.width,
window.height * l, window.width,
window.height), normalization);
cvCopy(img_feature_vector, &row);
j++;
}
for (int k = 0; k < horizontal_scans - 1; k++)
{
cvGetRow(training, &row, j);
img_feature_vector = calculateHOG_window(
integrals, cvRect(window.width * k,
img->height - window.height, window.width,
window.height), normalization);
cvCopy(img_feature_vector, &row);
j++;
}
cvGetRow(training, &row, j);
img_feature_vector = calculateHOG_window(integrals,
cvRect(img->width - window.width, img->height -
window.height, window.width, window.height),
normalization);
cvCopy(img_feature_vector, &row);
j++;
printf("%s\n", filename);
filename[prefix_length] = '\0';
for (int k = 0; k < 9; k++)
{
cvReleaseImage(&integrals[k]);
}
cvReleaseImage(&img);
}
printf("%d negative samples created \n",
training->rows);
if (savexml != NULL)
{
cvSave(savexml, training);
printf("Negative samples saved as %s\n",
savexml);
}
return training;
}
