Adaptive Labeled Multi-Bernoulli Filter with Pairwise Markov Chain Model and Student's t Noise

In the multitarget tracking (MTT) field, the MTT algorithm with hidden Markov chain (HMC) models typically assumes that process and measurement noises in the motion process obey independent Gaussian distributions. However, these assumptions of independence and Gaussianity do not always hold in many situations, such as, the tracking problem of noncooperative maneuvering targets with radar. As a result, this article proposes an adaptive labeled multi-Bernoulli (LMB) filter to handle the MTT problem when these assumptions of independence and Gaussianity are not satisfied. First, since the pairwise Markov chain (PMC) model's wider applicability compared to the HMC model and the Student's t distribution exhibits better heavy-tailed property than the Gaussian distribution, an MTT algorithm, abbreviated PMC-LMB-TM, is proposed by integrating the PMC model and the Student' s t mixture within the framework of the LMB filter. Among them, a Student' s t mixture matching method with Kullback-Leibler divergence (KLD) minimization is constructed to address the issue of the degree of freedom increase for the detecting targets during the updating process. Second, a KLD minimization-based adaptive estimation scheme for the PMC model is designed to address the problem with unknown noise scale matrices. Third, the proposed PMC-LMB-TM filter is combined with the proposed adaptive mechanism to construct a complete adaptive PMC-LMB-TM (PMC-LMB-ATM) algorithm for MTT problem with inaccurate noise scale matrices. Finally, the efficiency of the proposed algorithms is demonstrated through simulation experiments.