The effect of sensor registration error on the performance of PMHT

The probabilistic multi-hypothesis tracking (PMHT) algorithm has been successfully applied to a simulated multi-static active sonar data set that contains a single constant velocity target in varying amounts of clutter [1]. The simulated data set in that study contained negligible registration error was therefore easily registered to a common frame of reference for use in a centralized tracking architecture. Unknown sensor registration error can degrade the performance of any multi-static tracking algorithm that assumes that the measurement errors have zero mean. The registration error will appear as an unknown bias on the affected measurements and will degrade the data association stage of the tracker by increasing the probability that clutter detection will be confused for genuine target detections and vice versa. Moreover, if the registration error induced bias is large enough it may cause the tracking algorithm to misinterpret the track behavior (e.g., maneuvering instead of constant velocity). For many tracking algorithms this will cause the data association gates to be significantly expanded thereby increasing the probability that tracker will be drawn off target by clutter. The purpose of the effort reported here is to investigate the effect of unknown registration error on the tracking performance of PMHT in a centralized architecture. This study quantifies the ability of PMHT to maintain track on a maneuvering contact while it is in close proximity to a fixed persistent clutter object in the search region to using a simulated multi-static active sonar data set created by MSTWG researchers at TNO. Specifically, the probability of track hold as a function of target SNR and registration error between two receivers will be evaluated via simulation using the TNO data set. In this way estimates of the maximum amount of registration error that can be tolerated will be computed over a range of relevant target SNR values.