Consistency and robustness of PDAF for target tracking in cluttered environments

This paper presents a simulation-based investigation of the consistency of the PDAF via statistical tests of its actual errors versus the filter calculated covariances. These tests confirm the goodness of the approximation done at every stage in the PDAF, where a Gaussian mixture is replaced by an 'umbrella' Gaussian using moment matching. Two versions of the PDAF are examined: a parametric one that uses a Poisson model for the number of clutter originated measurements and a nonparametric one that uses a diffuse prior. While no new theoretical results are presented, the methodology of testing filter consistency can prove to be useful in the general evaluation of filters that contain approximations.