A novel adaptive δ‐generalised labelled multi‐Bernoulli filter for multi‐target tracking with heavy‐tailed noise

Abstract

Abstract Conventional δ‐generalised labelled multi‐Bernoulli filter (δ‐GLMB) cannot deal with the problem of the heavy‐tailed process noise and measurement noise. In order to solve this problem, an adaptive δ‐GLMB approach based on minimising Kullback‐Leibler Divergence (KLD) is proposed in this study. The inverse wishart and Student‐t mixture is used to approximate the joint posterior distribution of process noise covariance and measurement noise covariance together with multi‐target state, and the multi‐target state and noise parameters are jointly estimated by minimising the KLD. The simulation results show that the proposed approach is robust to multi‐target tracking for condition with heavy‐tailed process noise covariance and measurement noise covariance. Accurate target number and target state estimation are obtained by effectively estimating the process noise covariance and measurement noise covariance.

Keywords

• δ‐GLMB
• fixed iteration
• inverse Wishart
• KLD
• student‐t