Data association for GM-PHD with track oriented PMHT

Abstract

Gaussian Mixture probability hypothesis density (GM-PHD) filter is a closed-form solution to the probability hypothesis density filter, which could estimate states and time-varying number of targets based on theory of random finite set. Probability multiple hypotheses tracking (PMHT) is a multi-target tracking algorithm combining data association and expectation-maximization. However, GM-PHD can not give trajectories of target because of its disability of providing identity of target. Furthermore, PMHT need known number of targets and several frames trajectories of targets at first which are difficult in practical application. Firstly, we propose track oriented PMHT tracker (TO-PMHTT), then an approach of data association combining the advantage of GM-PHD with TO-PMHTT is designed in this paper. GM-PHD acts as the pre-filter of TO-PMHTT when there are no crossing targets in the scenario, while interaction between GM-PHD and TO-PMHTT is performed when targets enter crossing zone. Computer simulation results show that the method can provide association for both separated and crossing targets tracking.