A novel Gaussian-Student’s t-Skew mixture distribution based Kalman filter

Abstract

For the single-target tracking problem under multi-class noise mixing, the Gaussian-Student’s t-Skew mixture (GSTSM) distribution is proposed by introducing the Dirichlet random variables to model the mixed noise superimposed by multiple noise sources. By introducing multinomial random variables, the GSTSM distribution can be represented within a hierarchical model. This model is subsequently applied to the state–space model, employing a variational Bayesian (VB) approach to propose a novel robust Kalman filter based on the GSTSM distribution (GSTSM-KF). Simulation results show that GSTSM-KF can effectively improve the tracking accuracy in mixed noise scenarios.