Bayesian model selection for multisensor track-to-track association and track fusion

This paper derives a Bayesian procedure for track association that can solve a large scale distributed tracking problem where many sensors track many targets. When noninformative prior of the target state is assumed, the single target test becomes a chi-square test and it can be extended to the multiple target case by solving a multidimensional assignment problem. With the noninformative prior assumption, the optimal track fusion algorithm can be a biased one where the regularized estimate has smaller mean square estimation error. A regularized track fusion algorithm was presented which modifies the optimal linear unbiased fusion rule by a less-than-unity scalar. Simulation results indicate the effectiveness of the proposed track association and fusion algorithm through a three-sensor two-target tracking scenario