Abstract
In Bayesian multi-target tracking (MTT), knowledge of clutter intensity is required for effective multi-target state estimation. In this paper, we propose an online multi-target filter that can operate under background with unknown clutter intensity. Our solution is based on the Poisson multi-Bernoulli mixture (PMBM) filter that jointly estimating the multi-target state and clutter rate. The unknown clutter rate is modeled as Gamma distribution, consequently, the derived PMBM recursion that adapts for unknown clutter intensity remains closed. Moreover, we adopt a Gibbs sampler to find the finite number of global hypotheses, then the multi-Bernoulli mixture is approximated by a multi-Bernoulli distribution based on a simple fusion strategy. The derived Poison multi-Bernoulli (PMB) filter has a similar form with labeled multi-Bernoulli filter (LMB) but has a straightforward prediction step. Simulations conducted for linear-Gaussian models are presented to verify that the proposed algorithm can adapt to the background with unknown clutter intensity and yield reliable tracking performance.
Highlights
Multiple target tracking (MTT) is a hot topic with many important uses, for example, in aerospace applications, air traffic control, computer vision, surveillance and autonomous driving [1]–[3]
The filter is based on the Poisson multi-Bernoulli mixture (PMBM) filter
The unknown clutter rate is modeled as Gamma distribution, and a closed PMBM recursion for propagating the joint targets state and clutter rate is derived, along with an efficient implementation
Summary
Multiple target tracking (MTT) is a hot topic with many important uses, for example, in aerospace applications, air traffic control, computer vision, surveillance and autonomous driving [1]–[3]. These filters based on MTT conjugate priors have shown better tracking performance compared to PHD, cardinality PHD (CPHD) [26], [27] and MeMBer filters [9] Knowledge of parameters such as clutter rate and detection profile are of critical importance in Bayesian multi-target filtering. Recent works addressing the filtering problems with unknown clutter and detection parameters are based on the PHD, CPHD [28], cardinality balance MeMBer (CBMeMBer) [29] and δ-GLMB filter [30] These papers take the strategy proposed in [28], model the unknown detection probability as Beta distribution, and model individual clutter returns based on individual clutter generators.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have