Interval dominance based data association

A new robust filtering method has recently been proposed based on closed-convex sets of probability distributions or, equivalently, coherent lower previsions, which are used to characterize uncertainty in the prior, likelihood and, respectively, state transition models. In this paper, we generalize this approach to the multi-target tracking problem by also addressing the uncertainty on the origin of the measurements (target or clutter). In particular, we show that this further source of uncertainty can be taken into account by using set of distributions and decision techniques for coherent lower previsions. Finally, we evaluate the performance of the proposed tracker by means of Monte Carlo simulations relative to difficult tracking scenarios such as manoeuvring and crossing targets.