A robust and fast partitioning algorithm for extended target tracking using a Gaussian inverse Wishart PHD filter

Abstract

Extended target Gaussian inverse Wishart probability hypothesis density (ET-GIW-PHD) filter is a promising filter. However, the exact filter requires all possible partitions of the current measurement set for updating, which is computationally intractable. In order to limit the number of partitions, we propose a robust and fast partitioning algorithm, called modified Bayesian adaptive resonance theory (MB-ART) partition, based on Bayesian ART neural network architecture. In MB-ART partition, the alternative partitions approximating all possible partitions of the measurement set are generated by the different vigilance parameters, and these parameters are obtained by the bisection method. In addition, MB-ART partition can also solve the cardinality underestimation problem caused by the separating tracks scenario which was investigated by Granström et al. [1], since it takes into account the shape information of the different sized extended targets by iteratively updating variance. Simulation results show that our proposed partitioning algorithm can well handle the cardinality underestimation problem caused by the separating tracks scenario and reduce computational burden without losing tracking performance. For a four-target tracking scenario, the ET-GIW-PHD filter using MB-ART partition only requires 8.391 s on average for one Monte Carlo run, while the ET-GIW-PHD filter using combination partition requires 14.834 s. It implies that the proposed MB-ART partition has good application prospects for the real-time extended target tracking (ETT) system.