{
return cv::Size_<_Tp>(a.width / b, a.height / b);
}
+
+template<typename _Tp> static inline
+cv::Point_<_Tp> operator / (const cv::Point_<_Tp>& a, double b)
+{
+ return cv::Point_<_Tp>(a.x / b, a.y / b);
+}
+
#endif
class Kcf_Tracker_Private {
friend KCF_Tracker;
+
+ Kcf_Tracker_Private(const KCF_Tracker &kcf) : kcf(kcf) {}
+
+ const KCF_Tracker &kcf;
+#ifdef BIG_BATCH
std::vector<ThreadCtx> threadctxs;
+#else
+ ScaleRotVector<ThreadCtx> threadctxs{kcf.p_scales, kcf.p_angles};
+#endif
};
KCF_Tracker::KCF_Tracker(double padding, double kernel_sigma, double lambda, double interp_factor,
double output_sigma_factor, int cell_size)
: p_cell_size(cell_size), fft(*new FFT()), p_padding(padding), p_output_sigma_factor(output_sigma_factor), p_kernel_sigma(kernel_sigma),
- p_lambda(lambda), p_interp_factor(interp_factor), d(*new Kcf_Tracker_Private)
+ p_lambda(lambda), p_interp_factor(interp_factor)
{
}
-KCF_Tracker::KCF_Tracker() : fft(*new FFT()), d(*new Kcf_Tracker_Private) {}
+KCF_Tracker::KCF_Tracker() : fft(*new FFT()) {}
KCF_Tracker::~KCF_Tracker()
{
delete &fft;
- delete &d;
}
void KCF_Tracker::train(cv::Mat input_rgb, cv::Mat input_gray, double interp_factor)
// obtain a sub-window for training
get_features(input_rgb, input_gray, nullptr, p_current_center.x, p_current_center.y,
p_windows_size.width, p_windows_size.height,
- p_current_scale).copyTo(model->patch_feats.scale(0));
+ p_current_scale, p_current_angle).copyTo(model->patch_feats.scale(0));
DEBUG_PRINT(model->patch_feats);
fft.forward_window(model->patch_feats, model->xf, model->temp);
DEBUG_PRINTM(model->xf);
feature_size = fit_size / p_cell_size;
p_scales.clear();
- for (int i = -int(p_num_scales) / 2; i <= int(p_num_scales) / 2; ++i)
+ for (int i = -int(p_num_scales - 1) / 2; i <= int(p_num_scales) / 2; ++i)
p_scales.push_back(std::pow(p_scale_step, i));
+ p_angles.clear();
+ for (int i = -int(p_num_angles - 1) / 2; i <= int(p_num_angles) / 2; ++i)
+ p_angles.push_back(i * p_angle_step);
+
#ifdef CUFFT
if (m_use_linearkernel) {
std::cerr << "cuFFT supports only Gaussian kernel." << std::endl;
#endif
model.reset(new Model(feature_size, p_num_of_feats));
+ d.reset(new Kcf_Tracker_Private(*this));
#ifndef BIG_BATCH
for (auto scale: p_scales)
- d.threadctxs.emplace_back(feature_size, p_num_of_feats, scale);
+ for (auto angle : p_angles)
+ d->threadctxs.emplace_back(feature_size, p_num_of_feats, scale, angle);
#else
- d.threadctxs.emplace_back(feature_size, p_num_of_feats, p_num_scales);
+ d->threadctxs.emplace_back(feature_size, p_num_of_feats, p_scales, p_angles);
#endif
gaussian_correlation.reset(new GaussianCorrelation(1, p_num_of_feats, feature_size));
p_output_sigma = std::sqrt(p_init_pose.w * p_init_pose.h * double(fit_size.area()) / p_windows_size.area())
* p_output_sigma_factor / p_cell_size;
- fft.init(feature_size.width, feature_size.height, p_num_of_feats, p_num_scales);
+ fft.init(feature_size.width, feature_size.height, p_num_of_feats, p_num_scales * p_num_angles);
fft.set_window(MatDynMem(cosine_window_function(feature_size.width, feature_size.height)));
// window weights, i.e. labels
tmp.cy = p_current_center.y;
tmp.w = p_init_pose.w * p_current_scale;
tmp.h = p_init_pose.h * p_current_scale;
- tmp.a = 0;
+ tmp.a = p_current_angle;
if (p_resize_image)
tmp.scale(1 / p_downscale_factor);
}
}
+static void drawCross(cv::Mat &img, cv::Point center, bool green)
+{
+ cv::Scalar col = green ? cv::Scalar(0, 1, 0) : cv::Scalar(0, 0, 1);
+ cv::line(img, cv::Point(center.x, 0), cv::Point(center.x, img.size().height), col);
+ cv::line(img, cv::Point(0, center.y), cv::Point(img.size().height, center.y), col);
+}
+
+static cv::Point2d wrapAroundFreq(cv::Point2d pt, cv::Mat &resp_map)
+{
+ if (pt.y > resp_map.rows / 2) // wrap around to negative half-space of vertical axis
+ pt.y = pt.y - resp_map.rows;
+ if (pt.x > resp_map.cols / 2) // same for horizontal axis
+ pt.x = pt.x - resp_map.cols;
+ return pt;
+}
+
double KCF_Tracker::findMaxReponse(uint &max_idx, cv::Point2d &new_location) const
{
- double max = -1.;
- max_idx = std::numeric_limits<uint>::max();
+ double max;
+ const auto &vec = IF_BIG_BATCH(d->threadctxs[0].max, d->threadctxs);
#ifndef BIG_BATCH
- for (uint j = 0; j < d.threadctxs.size(); ++j) {
- if (d.threadctxs[j].max.response > max) {
- max = d.threadctxs[j].max.response;
- max_idx = j;
- }
- }
+ auto max_it = std::max_element(vec.begin(), vec.end(),
+ [](const ThreadCtx &a, const ThreadCtx &b)
+ { return a.max.response < b.max.response; });
#else
- for (uint j = 0; j < p_scales.size(); ++j) {
- if (d.threadctxs[0].max[j].response > max) {
- max = d.threadctxs[0].max[j].response;
- max_idx = j;
- }
- }
+ auto max_it = std::max_element(vec.begin(), vec.end(),
+ [](const ThreadCtx::Max &a, const ThreadCtx::Max &b)
+ { return a.response < b.response; });
#endif
- assert(max_idx < IF_BIG_BATCH(p_scales.size(), d.threadctxs.size()));
-
- if (m_visual_debug) {
- const bool rgb = true;
- int type = rgb ? d.threadctxs[0].dbg_patch[0].type() : d.threadctxs[0].response.type();
- int w = true ? 100 : (rgb ? fit_size.width : feature_size.width);
- int h = true ? 100 : (rgb ? fit_size.height : feature_size.height);
- cv::Mat all_responses(h * p_num_scales, w * p_num_angles, type, cv::Scalar::all(0));
- for (size_t i = 0; i < p_num_scales; ++i) {
- for (size_t j = 0; j < p_num_angles; ++j) {
- cv::Mat tmp;
- if (rgb) {
- tmp = d.threadctxs[IF_BIG_BATCH(0, p_num_angles * i + j)].dbg_patch[IF_BIG_BATCH(p_num_angles * i + j, 0)];
- } else {
- tmp = d.threadctxs[IF_BIG_BATCH(0, p_num_angles * i + j)].response.plane(IF_BIG_BATCH(p_num_angles * i + j, 0));
- tmp = circshift(tmp, -tmp.cols/2, -tmp.rows/2);
- }
- cv::resize(tmp, tmp, cv::Size(w, h));
- cv::Mat resp_roi(all_responses, cv::Rect(j * w, i * h, w, h));
- tmp.copyTo(resp_roi);
- }
- }
- cv::namedWindow("All responses", CV_WINDOW_AUTOSIZE);
- cv::imshow("All responses", all_responses);
- }
+ assert(max_it != vec.end());
+ max = max_it->IF_BIG_BATCH(response, max.response);
- cv::Point2i &max_response_pt = IF_BIG_BATCH(d.threadctxs[0].max[max_idx].loc, d.threadctxs[max_idx].max.loc);
- cv::Mat max_response_map = IF_BIG_BATCH(d.threadctxs[0].response.plane(max_idx), d.threadctxs[max_idx].response.plane(0));
+ max_idx = std::distance(vec.begin(), max_it);
+
+ cv::Point2i max_response_pt = IF_BIG_BATCH(max_it->loc, max_it->max.loc);
+ cv::Mat max_response_map = IF_BIG_BATCH(d->threadctxs[0].response.plane(max_idx),
+ max_it->response.plane(0));
DEBUG_PRINTM(max_response_map);
DEBUG_PRINT(max_response_pt);
- // sub pixel quadratic interpolation from neighbours
- if (max_response_pt.y > max_response_map.rows / 2) // wrap around to negative half-space of vertical axis
- max_response_pt.y = max_response_pt.y - max_response_map.rows;
- if (max_response_pt.x > max_response_map.cols / 2) // same for horizontal axis
- max_response_pt.x = max_response_pt.x - max_response_map.cols;
-
+ max_response_pt = wrapAroundFreq(max_response_pt, max_response_map);
+ // sub pixel quadratic interpolation from neighbours
if (m_use_subpixel_localization) {
new_location = sub_pixel_peak(max_response_pt, max_response_map);
} else {
new_location = max_response_pt;
}
DEBUG_PRINT(new_location);
+
+ if (m_visual_debug != vd::NONE) {
+ const bool fit = 1;
+ int w = fit ? 100 : (m_visual_debug == vd::PATCH ? fit_size.width : feature_size.width);
+ int h = fit ? 100 : (m_visual_debug == vd::PATCH ? fit_size.height : feature_size.height);
+ cv::Mat all_responses((h + 1) * p_num_scales - 1,
+ (w + 1) * p_num_angles - 1, CV_32FC3, cv::Scalar::all(0));
+ for (size_t i = 0; i < p_num_scales; ++i) {
+ for (size_t j = 0; j < p_num_angles; ++j) {
+ auto &threadctx = d->IF_BIG_BATCH(threadctxs[0], threadctxs(i, j));
+ cv::Mat tmp;
+ cv::Point2d cross = threadctx.IF_BIG_BATCH(max(i, j), max).loc;
+ cross = wrapAroundFreq(cross, max_response_map);
+ if (m_visual_debug == vd::PATCH ) {
+ threadctx.dbg_patch IF_BIG_BATCH((i, j),)
+ .convertTo(tmp, all_responses.type(), 1.0 / 255);
+ cross.x = cross.x / fit_size.width * tmp.cols + tmp.cols / 2;
+ cross.y = cross.y / fit_size.height * tmp.rows + tmp.rows / 2;
+ } else {
+ cv::cvtColor(threadctx.response.plane(IF_BIG_BATCH(threadctx.max.getIdx(i, j), 0)),
+ tmp, cv::COLOR_GRAY2BGR);
+ tmp /= max; // Normalize to 1
+ cross += cv::Point2d(tmp.size())/2;
+ tmp = circshift(tmp, -tmp.cols/2, -tmp.rows/2);
+ }
+ bool green = false;
+ if (&*max_it == &IF_BIG_BATCH(threadctx.max(i, j), threadctx)) {
+ // Show the green cross at position of sub-pixel interpolation (if enabled)
+ cross = new_location + cv::Point2d(tmp.size())/2;
+ green = true;
+ }
+ cross.x *= double(w)/tmp.cols;
+ cross.y *= double(h)/tmp.rows;
+ cv::resize(tmp, tmp, cv::Size(w, h));
+ drawCross(tmp, cross, green);
+ cv::Mat resp_roi(all_responses, cv::Rect(j * (w+1), i * (h+1), w, h));
+ tmp.copyTo(resp_roi);
+ }
+ }
+ cv::namedWindow("KCF visual debug", CV_WINDOW_AUTOSIZE);
+ cv::imshow("KCF visual debug", all_responses);
+ }
+
return max;
}
resizeImgs(input_rgb, input_gray);
#ifdef ASYNC
- for (auto &it : d.threadctxs)
+ for (auto &it : d->threadctxs)
it.async_res = std::async(std::launch::async, [this, &input_gray, &input_rgb, &it]() -> void {
it.track(*this, input_rgb, input_gray);
});
- for (auto const &it : d.threadctxs)
+ for (auto const &it : d->threadctxs)
it.async_res.wait();
#else // !ASYNC
NORMAL_OMP_PARALLEL_FOR
- for (uint i = 0; i < d.threadctxs.size(); ++i)
- d.threadctxs[i].track(*this, input_rgb, input_gray);
+ for (uint i = 0; i < d->threadctxs.size(); ++i)
+ d->threadctxs[i].track(*this, input_rgb, input_gray);
#endif
cv::Point2d new_location;
uint max_idx;
max_response = findMaxReponse(max_idx, new_location);
+ double angle_change = d->IF_BIG_BATCH(threadctxs[0].max, threadctxs).angle(max_idx);
+ p_current_angle += angle_change;
+
+ new_location.x = new_location.x * cos(-p_current_angle/180*M_PI) + new_location.y * sin(-p_current_angle/180*M_PI);
+ new_location.y = new_location.y * cos(-p_current_angle/180*M_PI) - new_location.x * sin(-p_current_angle/180*M_PI);
+
new_location.x *= double(p_windows_size.width) / fit_size.width;
new_location.y *= double(p_windows_size.height) / fit_size.height;
if (m_use_subgrid_scale) {
p_current_scale *= sub_grid_scale(max_idx);
} else {
- p_current_scale *= p_scales[max_idx];
+ p_current_scale *= d->IF_BIG_BATCH(threadctxs[0].max, threadctxs).scale(max_idx);
}
clamp2(p_current_scale, p_min_max_scale[0], p_min_max_scale[1]);
+
// train at newly estimated target position
train(input_rgb, input_gray, p_interp_factor);
}
TRACE("");
BIG_BATCH_OMP_PARALLEL_FOR
- for (uint i = 0; i < IF_BIG_BATCH(kcf.p_num_scales, 1); ++i)
+ for (uint i = 0; i < IF_BIG_BATCH(max.size(), 1); ++i)
{
- kcf.get_features(input_rgb, input_gray, &dbg_patch[i],
+ kcf.get_features(input_rgb, input_gray, &dbg_patch IF_BIG_BATCH([i],),
kcf.p_current_center.x, kcf.p_current_center.y,
kcf.p_windows_size.width, kcf.p_windows_size.height,
- kcf.p_current_scale * IF_BIG_BATCH(kcf.p_scales[i], scale))
+ kcf.p_current_scale * IF_BIG_BATCH(max.scale(i), scale),
+ kcf.p_current_angle + IF_BIG_BATCH(max.angle(i), angle))
.copyTo(patch_feats.scale(i));
DEBUG_PRINT(patch_feats.scale(i));
}
double min_val, max_val;
cv::Point2i min_loc, max_loc;
#ifdef BIG_BATCH
- for (size_t i = 0; i < kcf.p_scales.size(); ++i) {
+ for (size_t i = 0; i < max.size(); ++i) {
cv::minMaxLoc(response.plane(i), &min_val, &max_val, &min_loc, &max_loc);
DEBUG_PRINT(max_loc);
double weight = kcf.p_scales[i] < 1. ? kcf.p_scales[i] : 1. / kcf.p_scales[i];
// ****************************************************************************
cv::Mat KCF_Tracker::get_features(cv::Mat &input_rgb, cv::Mat &input_gray, cv::Mat *dbg_patch,
- int cx, int cy, int size_x, int size_y, double scale) const
+ int cx, int cy, int size_x, int size_y, double scale, double angle) const
{
cv::Size scaled = cv::Size(floor(size_x * scale), floor(size_y * scale));
- cv::Mat patch_gray = get_subwindow(input_gray, cx, cy, scaled.width, scaled.height);
- cv::Mat patch_rgb = get_subwindow(input_rgb, cx, cy, scaled.width, scaled.height);
+ cv::Mat patch_gray = get_subwindow(input_gray, cx, cy, scaled.width, scaled.height, angle);
+ cv::Mat patch_rgb = get_subwindow(input_rgb, cx, cy, scaled.width, scaled.height, angle);
if (dbg_patch)
patch_rgb.copyTo(*dbg_patch);
cv::Mat KCF_Tracker::circshift(const cv::Mat &patch, int x_rot, int y_rot) const
{
- cv::Mat rot_patch(patch.size(), CV_32FC1);
- cv::Mat tmp_x_rot(patch.size(), CV_32FC1);
+ cv::Mat rot_patch(patch.size(), patch.type());
+ cv::Mat tmp_x_rot(patch.size(), patch.type());
// circular rotate x-axis
if (x_rot < 0) {
// Returns sub-window of image input centered at [cx, cy] coordinates),
// with size [width, height]. If any pixels are outside of the image,
// they will replicate the values at the borders.
-cv::Mat KCF_Tracker::get_subwindow(const cv::Mat &input, int cx, int cy, int width, int height) const
+cv::Mat KCF_Tracker::get_subwindow(const cv::Mat &input, int cx, int cy, int width, int height, double angle) const
{
cv::Mat patch;
- int x1 = cx - width / 2;
- int y1 = cy - height / 2;
- int x2 = cx + width / 2;
- int y2 = cy + height / 2;
+ cv::Size sz(width, height);
+ cv::RotatedRect rr(cv::Point2f(cx, cy), sz, angle);
+ cv::Rect bb = rr.boundingRect();
+
+ int x1 = bb.tl().x;
+ int y1 = bb.tl().y;
+ int x2 = bb.br().x;
+ int y2 = bb.br().y;
// out of image
if (x1 >= input.cols || y1 >= input.rows || x2 < 0 || y2 < 0) {
// cv::waitKey();
}
+ cv::Point2f src_pts[4];
+ cv::RotatedRect(cv::Point2f(patch.size()) / 2.0, sz, angle).points(src_pts);
+ cv::Point2f dst_pts[3] = { cv::Point2f(0, height), cv::Point2f(0, 0), cv::Point2f(width, 0)};
+ auto rot = cv::getAffineTransform(src_pts, dst_pts);
+ cv::warpAffine(patch, patch, rot, sz);
+
// sanity check
assert(patch.cols == width && patch.rows == height);
return sub_peak;
}
-double KCF_Tracker::sub_grid_scale(uint index)
+double KCF_Tracker::sub_grid_scale(uint max_index)
{
cv::Mat A, fval;
- if (index >= p_scales.size()) {
+ const auto &vec = d->IF_BIG_BATCH(threadctxs[0].max, threadctxs);
+ uint index = vec.getScaleIdx(max_index);
+ uint angle_idx = vec.getAngleIdx(index);
+
+ if (index >= vec.size()) {
// interpolate from all values
// fit 1d quadratic function f(x) = a*x^2 + b*x + c
A.create(p_scales.size(), 3, CV_32FC1);
A.at<float>(i, 0) = float(p_scales[i] * p_scales[i]);
A.at<float>(i, 1) = float(p_scales[i]);
A.at<float>(i, 2) = 1;
- fval.at<float>(i) = d.threadctxs.back().IF_BIG_BATCH(max[i].response, max.response);
+ fval.at<float>(i) = d->IF_BIG_BATCH(threadctxs[0].max[i].response, threadctxs(i, angle_idx).max.response);
}
} else {
// only from neighbours
p_scales[index + 1] * p_scales[index + 1], p_scales[index + 1], 1);
#ifdef BIG_BATCH
fval = (cv::Mat_<float>(3, 1) <<
- d.threadctxs.back().max[index - 1].response,
- d.threadctxs.back().max[index + 0].response,
- d.threadctxs.back().max[index + 1].response);
+ d->threadctxs[0].max(index - 1, angle_idx).response,
+ d->threadctxs[0].max(index + 0, angle_idx).response,
+ d->threadctxs[0].max(index + 1, angle_idx).response);
#else
fval = (cv::Mat_<float>(3, 1) <<
- d.threadctxs[index - 1].max.response,
- d.threadctxs[index + 0].max.response,
- d.threadctxs[index + 1].max.response);
+ d->threadctxs(index - 1, angle_idx).max.response,
+ d->threadctxs(index + 0, angle_idx).max.response,
+ d->threadctxs(index + 1, angle_idx).max.response);
#endif
}
projects / hercules2020 / kcf.git / blobdiff