A Gaussian Uniform Mixture Model for Robust Kalman Filtering

Abstract

This article presents a Kalman-type recursive estimator for discrete-time systems with a measurement noise modeled by a Gaussian-uniform mixture. The objective is to deal with data containing outliers that degrade the performance of the regular Kalman filter. The proposed non-Gaussian noise model takes into account the reliability of the measurement with respect to erroneous data. The Kalman-type estimator is based on Masreliez's formulation which copes with non-Gaussian noise models. Results in different simulated conditions are displayed to evaluate the performance of the newly-presented algorithm and to compare it to state-of-art alternatives.