Distributed multi‐sensor fusion using generalised multi‐Bernoulli densities

Abstract

A new method for distributed multi-target tracking with multistatic radar systems is presented.The proposed method is based on using generalised covariance intersection (GCI) of multi-object densities for fusion of the posteriors within a multi-object Bayesian filtering scheme. The presented solution is particularly formulated for sensor fusion with posterior densities that are parameterised as generalised multi-Bernoulli (GMB) distributions which are the unlabelled version of VoCVo densities by discarding the labels. To obtain a closed-form solution for fusing GMB densities, the authors use an efficient approximation to the densities. The approximated density is another GMB density that preserves both the first-order moment (intensity or PHD) and the cardinality distribution of the original density. As such, it is called the second-order approximation of the GMB (SO-GMB) density. The resulting explicit expressions for the GCI fusion using SO-GMB approximations allow distributed sensor fusion, not only with VoCVo filters, but also with M-generalised labelled multi-Bernoulli and labelled multi-Bernoulli filters being in place as local filters in the multistatic radar system. In two challenging multi-target tracking scenarios, the tracking performance of the proposed method is shown to outperform the state of art.