Poisson Multi-Bernoulli Mixture Filter for Multiple Extended Object Tracking Using Kolmogorov–Smirnov Test

Abstract

The Poisson multi-Bernoulli mixture (PMBM) filter has been proved to be effective in tracking missed objects and an unknown number of objects with clutter in complex scenarios. However, when the objects are closely spaced, spawning or maneuvering will affect the performance of traditional object prior-based partitioning methods. To solve this problem, this article proposes an expectation-maximization (EM) partitioning method for PMBM filtering of extended target tracking based on the improved Kolmogorov-Smirnov (KS) test. The initial partitioning and correlation of measurements are performed from the perspective of measurement distribution. Data association is further optimized by combining with the prior to obtain more reasonable associations and guide the generation of the spawning object Poisson point process (PPP). In addition, a method is proposed for dynamically generating the spawning object PPP to achieve accurate tracking of spawning objects. Simulation results show that the proposed method has good robustness in scenarios involving intersection, closely spaced objects, and derivation compared with the Gamma-Gaussian inverse Wishart (GGIW)-PMBM filter.