MT::MT(const unsigned char *gray, int width, int height, rect_t rect, ostream *os) :
image_width(width), image_height(height),
window_width(rect.width), window_height(rect.height),
warp(width, height),
feature(width, height),
log(os)
{
//build the fine template on uniform grid
warp.sett(locate(rect));
float fine_stride = sqrt(window_width * window_height / fine_n);
int W = int(floor(window_width / (2.0f * fine_stride)));
int H = int(floor(window_height / (2.0f * fine_stride)));
for (int y = 0; y <= 2 * H; ++y)
for (int x = 0; x <= 2 * W; ++x)
fine_samples.push_back(Vector3f((x - W) * fine_stride, (y - H) * fine_stride, 0.0f));
//initialization
feature.process(gray, 0.0f);
N = 0;
fine_train(warp);
fast_train(warp);
fine_errors.push_back(0.0f);
error = 0.0f;
roll = yaw = pitch = 0.0f;
number_coarse = number_MLK = number_iteration = 0;
}
Rect2f MT::track(Mat img, const vector<Rect2f> &detections)
{
if (log != NULL) {
(*log) << "roll = " << roll * 90.0f / PI_2 << endl;
(*log) << "yaw = " << yaw * 90.0f / PI_2 << endl;
(*log) << "pitch = " << pitch * 90.0f / PI_2 << endl;
}
if (!detections.empty())
roll = 0.0f;
feature.process(img, roll);
vector<Point3f> candidates;
candidates.push_back(warp.t);
candidates.push_back(fast_test(warp));
for (auto d : detections)
candidates.push_back(locate(d));
Warp best_warp(image_size);
float best_error = 1.0f;
for (auto& t : candidates) {
if (log != NULL)
(*log) << "candidate " << t << " " << window(t) << endl;
Warp w(image_size);
if (detections.empty())
w.set(warp.r);
w.set(t);
w = fine_test(w);
float e = evaluate(w);
if (log != NULL) {
(*log) << "final translation = " << w.t << " " << window(w.t) << endl;
(*log) << "final rotation = " << w.r << endl;
(*log) << "final error = " << e << endl;
}
if (e < best_error) {
best_warp = w;
best_error = e;
}
}
candidates.clear();
error = best_error;
if (error < threshold_error) {
count = 0;
warp = best_warp;
warp.euler(roll, yaw, pitch);
fine_train(warp);
}
else {
++count;
warp.t = best_warp.t * (warp.t.z / best_warp.t.z);
}
fast_train(warp);
return window(best_warp.t);
}
void MT::update(Warp w, float e)
{
if (log != NULL) {
(*log) << "final translation = " << w.t.transpose() << " " << window(w.t) << endl;
(*log) << "final rotation = " << w.r.transpose() << endl;
(*log) << "final error = " << e << endl;
}
//if the error is lower than threshold, we update the fine model
error = e;
if (e < fine_threshold) {
warp = w;
warp.euler(roll, yaw, pitch);
fine_train(warp);
}
else
warp.t = w.t * (warp.t.z() / w.t.z());
fast_train(warp);
fine_errors.push_back(e);
while (fine_errors.size() > detect_interval)
fine_errors.pop_front();
}
MT::MT(Mat img, Rect2f rect, ostream *os) :
log(os),
image_size(img.size()),
window_size(rect.size()),
feature(image_size),
warp(image_size)
{
warp.set(locate(rect));
float fine_stride = sqrt(window_size.area() / fine_n);
int W = int(floor(window_size.width / (2.0f * fine_stride)));
int H = int(floor(window_size.height / (2.0f * fine_stride)));
for (int y = 0; y <= 2 * H; ++y)
for (int x = 0; x <= 2 * W; ++x)
fine_samples.push_back(Point3f((x - W) * fine_stride, (y - H) * fine_stride, 0.0f));
feature.process(img, 0.0f);
fine_train(warp);
fast_train(warp);
error = 0.0f;
roll = yaw = pitch = 0.0f;
count = N = 0;
}
