Poisson Multi-Bernoulli Approximations for Multiple Extended Object Filtering

The Poisson multi-Bernoulli mixture (PMBM) is a multiobject conjugate prior for the closed-form Bayes random finite set filter. extended object PMBM filter provides solution multiple filtering with standard models. This article considers computationally lighter alternatives to by propagating (PMB) density through recursion. A new local hypothesis representation presented, where each measurement creates Bernoulli component. facilitates developments of methods efficiently approximating posterior after update step as PMB. Based on representation, two approximation are presented: one based track-oriented (MB) approximation, and other variational MB via Kullback–Leibler divergence minimization. performance proposed PMB filters gamma Gaussian inverse-Wishart implementations evaluated in simulation study.