Bayesian Sequential Track Formation

Abstract

This paper presents a theoretical framework for track building in multiple-target scenarios from the Bayesian point of view. It is assumed that the number of targets is fixed and known. We propose two optimal methods for building tracks sequentially. The first one uses the labelling of the current multitarget state estimate that minimizes the mean-square labeled optimal subpattern assignment error. This method requires knowledge of the posterior density of the vector-valued state. The second assigns the labeling that maximizes the probability that the current multi-target state estimate is optimally linked with the available tracks at the previous time step. In this case, we only require knowledge of the random finite-set posterior density without labels.