Making the probabilistic multi-hypothesis tracker the tracker of choice

The probabilistic multi-hypothesis tracker (PMHT) is an algorithm of considerable beauty. In practice, its performance turns out to be similar to that of the probabilistic data association filter (PDAF) in many cases; and since the implementation of the PDAF is less intense numerically the PMHT has been having a hard time finding acceptance. The task, therefore, is to "make the PMHT work". The PMHT's problems are expressed as "nonadaptivity", "narcissism", and "over-hospitality to clutter". In this paper we show (in section 3) modifications to the original basic PMHT which offer some improvement. We explore, among other things, "homothetic" measurement models; "endpoint" and "start-point" PMHTs, which modify the PMHT assumptions such that the estimation goal is the track at only one point; maneuver-based PMHTs, including those with separate and joint homothetic measurement models; a modified PMHT whose measurement/target association model is more similar to that of the PDAF; and PMHTs with eccentric and/or estimated measurement models. The above are improvements on the basic PMHT based on modifications of the underlying model. We also offer several basic-PMHT implementations with improved convergence properties.