Comparison of PMHT and S-D assignment trackers

Probably the most popular extant multitarget/multisensor tracking algorithms are variations on one of the joint probabilistic data-association filter (JPDAF), the multi-hypothesis tracker (MHT), the S-dimensional assignment (SDA) tracker, and/or the probabilistic multi-hypothesis tracker (PMHT). The two better-known former algorithms suffer when any or all of the number of targets, of sensors, or of false alarms becomes large; this is due to their exponential increase in complexity as a function of most of these parameters. The computational load in the latter two schemes is considerably lower: generally polynomial in the SDA, and even linear in the PMHT, as a function of the number of targets. Because of this, and because of its good performance, SDA is becoming a standard tracking approach; the PMHT is somewhat newer and has yet to prove itself. In this paper we compare the SDA and PMHT, both for multi-sensor (data fusion) and dense multitarget situations. Both approaches perform quite well in terms of estimation, hence the comparison is fundamentally in term of computational load. It is observed that for extremely challenging environments the PMHT is favored