A robust distributed interaction multiple model filter for jump Markov systems with heavy-tailed measurement noise

Abstract

When heavy-tailed measurement noises (HTMNs) are introduced by outliers, the estimation accuracy of traditional distributed interaction multiple model (IMM) algorithms will be severely reduced in some target tracking scenarios. To solve the problem, a new distributed robust filter algorithm using cubature information filtering and weighted average consensus approach in the framework of IMM filtering is proposed. Firstly, the HTMNs are modeled as Student’s t-distributions, and as conjugate prior distributions for the degree of freedom parameters and the scale matrices, the Gamma distributions and the inverse-Wishart distributions have been chosen. Secondly, the system state vectors, noise corresponding parameters, and model probabilities are inferred utilizing the variational Bayesian method. What’s more, in order to increase the stability of sensor networks and improve the estimation accuracy of the proposed filtering algorithm, the weighted average consensus approach is used to update information pairs and model probabilities. Finally, a target tracking simulation experiment is used to verify that the proposed algorithm has higher estimation accuracy and robustness than existing advanced filtering algorithms in dealing with HTMN.