Joint tracking and classification based on kinematic and target extent measurements

A great deal of interest has been paid to target tracking for the last decades. When using Bayesian estimation algorithms, choosing relevant motion models is crucial for accurate localization. Information on the type of target and its maneuver capability can be helpful in the motion model design. Thus, joint tracking and classification (JTC) methods based on target features have been recently developed. In this paper, JTC is addressed by using target extent measurements. We present a flexible formulation of the JTC problem where a target class is characterized by a set of possible motion models. Two multiclass multiple-model algorithms are first derived. Then, to alleviate the difficult tuning of the model parameters, we take advantage of Bayesian non-parametric models. A Dirichlet-process based algorithm is presented for the JTC and the model parameter estimation. Finally, a comparative study of these three approaches is carried out for maritime-target tracking.