A fast Poisson labeled multi-Bernoulli filter for extended object tracking using belief propagation

Abstract

This paper addresses the multiple extended object tracking problem to enhance tracking accuracy and efficiency while ensuring track continuity. We propose a novel parameter-based PLMB-BP filter that integrates random finite set (RFS) and belief propagation (BP) methods. Poisson and labeled multi-Bernoulli (PLMB) RFSs are employed to model the states of newborn objects and multiple extended objects. By leveraging their advantages, the proposed filter simultaneously ensures track continuity and enhances birth model flexibility. Furthermore, the parameter-based BP is implemented for the marginal probability density function of object states and association variables. Inspired by fixed-point iteration, this implementation achieves joint estimation of measurement rate, kinematic state, and extent state for multiple extended objects, while maintaining superior real-time capability. Simulations are performed for closely spaced multiple extended objects with ellipsoidal shapes. The results demonstrate the enhanced tracking performance and the superior real-time capability of the proposed filter.