#include "linear_assignment.h"
|
#include "../Munkres_assign/hungarianoper.h"
|
#include <map>
|
|
linear_assignment *linear_assignment::instance = NULL;
|
linear_assignment::linear_assignment()
|
{
|
}
|
|
linear_assignment *linear_assignment::getInstance()
|
{
|
if(instance == NULL) instance = new linear_assignment();
|
return instance;
|
}
|
|
linear_assignment::~linear_assignment()
|
{
|
delete instance;
|
}
|
|
TRACHER_MATCHD
|
linear_assignment::matching_cascade(
|
tracker *distance_metric,
|
tracker::GATED_METRIC_FUNC distance_metric_func,
|
float max_distance,
|
int cascade_depth,
|
std::vector<std::shared_ptr<Track>> &tracks,
|
const DETECTIONS &detections,
|
std::vector<int>& track_indices,
|
std::vector<int> detection_indices)
|
{
|
TRACHER_MATCHD res;
|
for(size_t i = 0; i < detections.size(); i++) {
|
detection_indices.push_back(int(i));
|
}
|
|
std::vector<int> unmatched_detections;
|
unmatched_detections.assign(detection_indices.begin(), detection_indices.end());
|
res.matches.clear();
|
std::vector<int> track_indices_l;
|
|
std::map<int, int> matches_trackid;
|
for(int level = 0; level < cascade_depth; level++) {
|
if(unmatched_detections.size() == 0) break; //No detections left;
|
|
track_indices_l.clear();
|
for(int k:track_indices) {
|
if(tracks[k]->time_since_update == 1+level)
|
track_indices_l.push_back(k);
|
}
|
if(track_indices_l.size() == 0) continue; //Nothing to match at this level.
|
|
TRACHER_MATCHD tmp = min_cost_matching(
|
distance_metric, distance_metric_func,
|
max_distance, tracks, detections, track_indices_l,
|
unmatched_detections);
|
unmatched_detections.assign(tmp.unmatched_detections.begin(), tmp.unmatched_detections.end());
|
for(size_t i = 0; i < tmp.matches.size(); i++) {
|
MATCH_DATA pa = tmp.matches[i];
|
res.matches.push_back(pa);
|
matches_trackid.insert(pa);
|
}
|
}
|
res.unmatched_detections.assign(unmatched_detections.begin(), unmatched_detections.end());
|
|
for(size_t i = 0; i < track_indices.size(); i++) {
|
int tid = track_indices[i];
|
if(matches_trackid.find(tid) == matches_trackid.end())
|
res.unmatched_tracks.push_back(tid); // 未匹配的追加
|
}
|
|
return res;
|
}
|
|
TRACHER_MATCHD
|
linear_assignment::min_cost_matching(tracker *distance_metric,
|
tracker::GATED_METRIC_FUNC distance_metric_func,
|
float max_distance,
|
std::vector<std::shared_ptr<Track>> &tracks,
|
const DETECTIONS &detections,
|
std::vector<int> &track_indices,
|
std::vector<int> &detection_indices)
|
{
|
TRACHER_MATCHD res;
|
if((detection_indices.size() == 0) || (track_indices.size() == 0)) {
|
res.matches.clear();
|
res.unmatched_tracks.assign(track_indices.begin(), track_indices.end());
|
res.unmatched_detections.assign(detection_indices.begin(), detection_indices.end());
|
return res;
|
}
|
DYNAMICM cost_matrix = (distance_metric->*(distance_metric_func))( // distance_metric_func其实是调用tracker.cpp中的iou_cost 函数
|
tracks, detections, track_indices, detection_indices);
|
for(int i = 0; i < cost_matrix.rows(); i++) {
|
for(int j = 0; j < cost_matrix.cols(); j++) {
|
float tmp = cost_matrix(i,j);
|
if(tmp > max_distance) cost_matrix(i,j) = max_distance + 1e-5;
|
}
|
}
|
Eigen::Matrix<float, -1, 2, Eigen::RowMajor> indices = HungarianOper::Solve(cost_matrix);
|
res.matches.clear();
|
res.unmatched_tracks.clear();
|
res.unmatched_detections.clear();
|
for(size_t col = 0; col < detection_indices.size(); col++) {
|
bool flag = false;
|
for(int i = 0; i < indices.rows(); i++)
|
if(indices(i, 1) == col) { flag = true; break;}
|
if(flag == false)res.unmatched_detections.push_back(detection_indices[col]);
|
}
|
for(size_t row = 0; row < track_indices.size(); row++) {
|
bool flag = false;
|
for(int i = 0; i < indices.rows(); i++)
|
if(indices(i, 0) == row) { flag = true; break; }
|
if(flag == false) res.unmatched_tracks.push_back(track_indices[row]);
|
}
|
for(int i = 0; i < indices.rows(); i++) {
|
int row = indices(i, 0);
|
int col = indices(i, 1);
|
|
int track_idx = track_indices[row];
|
int detection_idx = detection_indices[col];
|
if(cost_matrix(row, col) > max_distance) {
|
res.unmatched_tracks.push_back(track_idx);
|
res.unmatched_detections.push_back(detection_idx);
|
} else res.matches.push_back(std::make_pair(track_idx, detection_idx));
|
}
|
return res;
|
}
|
|
DYNAMICM
|
linear_assignment::gate_cost_matrix(
|
KalmFilter *kf,
|
DYNAMICM &cost_matrix,
|
std::vector<std::shared_ptr<Track>> &tracks,
|
const DETECTIONS &detections,
|
const std::vector<int> &track_indices,
|
const std::vector<int> &detection_indices,
|
float gated_cost, bool only_position)
|
{
|
int gating_dim = (only_position == true?2:4);
|
double gating_threshold = KalmFilter::chi2inv95[gating_dim];
|
std::vector<DETECTBOX> measurements;
|
for(int i:detection_indices)
|
{
|
DETECTION_ROW t = detections[i];
|
measurements.push_back(t.to_xyah());
|
}
|
|
for(size_t i = 0; i < track_indices.size(); i++)
|
{
|
auto& track = tracks[track_indices[i]];
|
Eigen::Matrix<float, 1, -1> gating_distance = kf->gating_distance(
|
track->mean, track->covariance, measurements, only_position);
|
|
for (int j = 0; j < gating_distance.cols(); j++)
|
{
|
if (gating_distance(0, j) > gating_threshold) cost_matrix(i, j) = gated_cost;
|
}
|
}
|
return cost_matrix;
|
}
|
|
