Poisson multi-bernoulli mixture filter for enhanced tracking of spawning targets

Abstract

This paper introduces an enhanced Poisson multi-Bernoulli mixture (PMBM) filter for spawning targets, wherein spawning refers to the separation of one or multiple objects from an existing target. Tracking such targets poses a significant challenge due to the unknown location at which a target may spawn. Leveraging the information offered by the density of existing group targets, the proposed PMBM filter enables the prediction of spawning for all members. Through modifications based on the latest state of detected group targets in the Bernoulli components, the detection probability for spawning is enhanced, consequently reducing the error stemming from missed targets. This approach yields a favorable trade-off in computational complexity by modeling spawning through the Poisson Random Finite Set (RFS) in the filter, thereby averting the generation of Bernoulli components for spawned and undetected group targets. Monte Carlo simulations indicate that the modified PMBM filter diminishes missed targets and false alarms while enhancing tracking reliability during target spawning events.