Tracking and classifying multiple targets without<tex>a priori</tex>identification

Abstract

Based on a general target/sensor model which allows dependence among targets and state-dependent target detection, a Bayesian solution to the multitarget multisensor tracking problem is derived for cases where targets do not have a priori identification, i.e., targets are not labeled a priori. When this solution is applied to a class of independent target models, a more implementable class of algorithms is obtained. A clear definition is given to a newly-detected-target likelihood, thereby eliminating the ambiguous notion of Poisson arrival of new targets. Representative existing algorithms are then compared to our results.