Application of robust filtering and smoothing to tracking data

Abstract

Robust methods provide a fresh approach to the problem of treatment of wild observations in filtering and smoothing problems. The robust M-estimates of regression are extended to filtering and fixed lag smoothing employing a pseudodensity of the observations in a maximum likelihood derivation of the filter and fixed lag smoother. These robust methods have been applied to simulated and real tracking data to obtain improved estimstion performance in the presence of wild observations.