void prepareMeasurement(const ed::EntityConstPtr& e, cv::Mat& view_color_img, cv::Mat& view_depth_img, cv::Mat& mask, cv::Rect& bouding_box) {
// Get the best measurement from the entity
ed::MeasurementConstPtr msr = e->lastMeasurement();
if (!msr)
return;
int min_x, max_x, min_y, max_y;
// create a view
rgbd::View view(*msr->image(), msr->image()->getRGBImage().cols);
// get depth image
view_depth_img = msr->image()->getDepthImage();
// get color image
const cv::Mat& color_image = msr->image()->getRGBImage();
// crop it to match the view
view_color_img = cv::Mat(color_image(cv::Rect(0, 0, view.getWidth(), view.getHeight())));
// std::cout << "image: " << cropped_image.cols << "x" << cropped_image.rows << std::endl;
// initialize bounding box points
max_x = 0;
max_y = 0;
min_x = view.getWidth();
min_y = view.getHeight();
// initialize mask, all 0s
mask = cv::Mat::zeros(view.getHeight(), view.getWidth(), CV_8UC1);
// Iterate over all points in the mask
for(ed::ImageMask::const_iterator it = msr->imageMask().begin(view.getWidth()); it != msr->imageMask().end(); ++it)
{
// mask's (x, y) coordinate in the depth image
const cv::Point2i& p_2d = *it;
// paint a mask
mask.at<unsigned char>(*it) = 255;
// update the boundary coordinates
if (min_x > p_2d.x) min_x = p_2d.x;
if (max_x < p_2d.x) max_x = p_2d.x;
if (min_y > p_2d.y) min_y = p_2d.y;
if (max_y < p_2d.y) max_y = p_2d.y;
}
bouding_box = cv::Rect(min_x, min_y, max_x - min_x, max_y - min_y);
}
void HumanContourMatcher::process(ed::EntityConstPtr e, tue::Configuration& result) const
{
// if initialization failed, return
if (!init_success_)
return;
// Get the best measurement from the entity
ed::MeasurementConstPtr msr = e->lastMeasurement();
if (!msr)
return;
float depth_sum = 0;
float avg_depht;
float classification_error = 0;
float classification_deviation = 0;
std::string classification_stance;
uint point_counter = 0;
bool is_human = false;
// Get the depth image from the measurement
const cv::Mat& depth_image = msr->image()->getDepthImage();
const cv::Mat& color_image = msr->image()->getRGBImage();
cv::Mat mask_cv = cv::Mat::zeros(depth_image.rows, depth_image.cols, CV_8UC1);
// Iterate over all points in the mask. You must specify the width of the image on which you
// want to apply the mask (see the begin(...) method).
for(ed::ImageMask::const_iterator it = msr->imageMask().begin(depth_image.cols); it != msr->imageMask().end(); ++it)
{
// mask's (x, y) coordinate in the depth image
const cv::Point2i& p_2d = *it;
// TODO dont creat a Mat mask, just give human_classifier_.Classify a vector of 2D points!
// paint a mask
mask_cv.at<unsigned char>(p_2d) = 255;
// calculate measurement average depth
depth_sum += depth_image.at<float>(p_2d);
point_counter++;
}
avg_depht = depth_sum/(float)point_counter;
is_human = human_classifier_.Classify(depth_image, color_image, mask_cv, avg_depht, classification_error, classification_deviation, classification_stance);
// ----------------------- assert results -----------------------
// create group if it doesnt exist
if (!result.readGroup("perception_result", tue::OPTIONAL))
{
result.writeGroup("perception_result");
}
// assert the results to the entity
result.writeGroup(kModuleName);
result.setValue("label", "Human Shape");
if (classification_error > 0){
result.setValue("stance", classification_stance);
result.setValue("error", classification_error);
result.setValue("deviation", classification_deviation);
}
if(is_human)
{
result.setValue("score", 1.0);
}else{
result.setValue("score", 0.0);
}
result.endGroup(); // close human_contour_matcher group
result.endGroup(); // close perception_result group
}