Particle filter for extracting target label information when targets move in close proximity

Abstract

This paper addresses the problem of approximating the posterior probability density function of two targets after a crossing from the Bayesian perspective such that the information about target labels is not lost. To this end, we develop a particle filter that is able to maintain the inherent multimodality of the posterior after the targets have moved in close proximity. Having this approximation available, we are able to extract information about target labels even when the measurements do not provide information about target's identities. In addition, due to the structure of our particle filter, we are able to use an estimator that provides lower optimal subpattern assignment (OSPA) errors than usual estimators.