Adaptive Poisson multi-Bernoulli filter for multiple extended targets with Gamma and Beta estimator

Cheng Chen, Jinlong Yang, Jianjun Liu

Abstract: The Poisson multi-Bernoulli (PMB) filter has been proven to be an effective method for multiple target tracking (MTT), however, some parameters such as clutter rate and detection probability are usually unknown in practical tracking scenarios, which can affect the tracking accuracy of the algorithm. To solve this problem, we propose a robust Poisson multi-Bernoulli filter with independent clutter rate estimator and detection probability estimator, referred to as GBePMB, which can online estimate the unknown parameters for extended target tracking (ETT) scenario. The closed-form solution to the clutter rate estimator is derived by using the maximum likelihood estimation (MLE) technique and Gamma conjugate prior. The detection probability estimator uses the Beta distribution to describe the unknown detection probability, and the Beta variational approximation is proposed to adapt to the iterative requirements of PMB. Finally, simulation results show that the proposed algorithm has a good performance and robustness under unknown clutter rate and detection probability.