A multiple model multiple hypothesis filter for systems with possibly erroneous measurements

In this paper a novel method to deal with possibly erroneous measurements is presented. In target tracking applications it may be the case that measurements that are obtained are incorrect in the sense that they do not comply with the measurement model. Examples in (radar) target tracking are: Glint, Multipath, Ambiguous Doppler, etc. The method that we present here is able to detect these non-normalities and modifies the measurement model in such a way that these non-normalities do not blur the track filter output. The method is based on a multi hypothesis assumption w.r.t. to the correctness of the measurement model. This new method is also shown to outperform classical methods for dealing with possibly erroneous measurements. We demonstrate our method by an extensive example of a surveillance radar tracking system with unreliable (or sometimes false) Doppler measurements.