Event

GenTrack3 proposes hybrid stochastic-deterministic online multi-object tracking framework

1 reports · 1 sources

Toan Van Nguyen and other authors propose GenTrack3, an online multi-object tracking (MOT) framework that integrates deterministic and stochastic principles. The framework uses stochastic particle sets to compute matching costs between targets and detections, which the authors claim maintains tracking robustness under uncertainty; the paper also introduces a new track-detection matching method, which they claim improves scalability as the number of targets increases and supports group tracking. Tracking inference uses tracklets containing identity, state, velocity, track penalty, and track age. The authors state that the reference implementation and baseline trackers have been released on GitHub; the paper content has been incorporated into the complete GenTrack series manuscript submitted to a journal.