Improving Tracking of Splitting Group Targets Using the Main Target Density in the PMBM Filter

The Poisson Multi-Bernoulli Mixture filter is one of the most efficient filters for group target tracking. In this filter, target spawning, i.e., the appearance of a new target in the proximity of an existing one in the surveillance area is modeled as a newborn group target. Using this approach may result in missed targets or false alarms. In this paper, profiting from useful information provided by the density of existing group targets, it is possible to predict spawning for all members in the surveillance area. With modification in the birth model in the Poisson density of the filter based on the latest state of detected group targets in the Bernoulli part, the spawning detection probability increases, and the error caused by missed targets is reduced. This approach benefits from the moderated computational complexity property of this filter, particularly for splitting group/point targets, and prevents generating new Bernoulli components for spawned and undetected group targets. The results of Monte Carlo simulations confirm that the modified Poisson Multi-Bernoulli Mixture filter can reduce missed targets and false alarms and increase the reliability of tracking.