Track Coalescence and Repulsion: MHT, JPDA, and BP

TL;DR

This paper reviews and compares MHT, JPDA, and BP methods for multitarget tracking, revealing that BP can largely avoid the issues of track coalescence and repulsion seen in traditional methods.

Contribution

It introduces a comprehensive comparison of estimation strategies and demonstrates BP's ability to mitigate common tracking problems.

Findings

BP-based MTT largely avoids track coalescence and repulsion

JPDA suffers from track coalescence in close target scenarios

MHT exhibits track repulsion, opposite to coalescence

Abstract

Joint probabilistic data association (JPDA) and multiple hypothesis tracking (MHT) introduced in the 70s, are still widely used methods for multitarget tracking (MTT). Extensive studies over the last few decades have revealed undesirable behavior of JPDA and MHT type methods in tracking scenarios with targets in close proximity. In particular, JPDA suffers from the track coalescence effect, i.e., estimated tracks of targets in close proximity tend to merge and can become indistinguishable. Interestingly, in MHT, an opposite effect to track coalescence called track repulsion can be observed. In this paper, we review the estimation strategies of the MHT, JPDA, and the recently introduced belief propagation (BP) framework for MTT and we investigate if BP also suffers from these two detrimental effects. Our numerical results indicate that BP-based MTT can mostly avoid both track repulsion and coalescence.