A Robust Particle Filtering Algorithm With Non-Gaussian Measurement Noise Using Student-t Distribution

The Gaussian noise assumption may result in a major decline in state estimation accuracy when the measurements are with the presence of outliers. In this letter, we endow the unknown measurement noise with the Student-t distribution to model the underlying non-Gaussian dynamics of a real physical system. Thereafter a robust particle filtering algorithm is developed. First, we employ variational Bayesian (VB) approach to robustly infer the unknown noise parameters recursively. Second, in order to decrease the computational complexity resulted by the unknown noise parameters, those parameters are marginalized out to allow each particle to be updated by using sufficient statistics estimated by VB approach. The proposed algorithm is tested with a typical non-linear model and the robustness of our algorithm has been borne out.