Robust adaptive multi-target tracking with unknown heavy-tailed noise

In multi-target tracking, non-Gaussian heavy-tailed process noise (PN) and measurement noise (MN) are introduced by unknown manoeuvring and noise-corrupted measurements. This study proposes a Gaussian approximation approach based on multivariate Student-t distribution, which is designed to characterise non-Gaussian heavy-tailed MN covariance and PN covariance. The variational Bayesian approach is applied to a generalised labelled multi-Bernoulli (GLMB) with an augmented state, and a robust adaptive generalised labelled multi-Bernoulli (RAGLMB) framework is derived to recursively propagate the joint posterior density of noise covariance and target state. The simulation results indicate that the proposed RAGLMB filter is robust to targets affected by non-Gaussian heavy-tailed PN and MN.