Variational Bayesian-Based Maximum Correntropy Cubature Kalman Filter With Both Adaptivity and Robustness

Abstract

This paper focuses on solving the problems of unknown measurement noise covariance and measurement outliers, which occurs in the vision/dual-IMU integrated attitude determination system. Although many adaptive filters and robust filters have been proposed to deal with the unknown measurement noise covariance or measurement outliers, most of them cannot handle both the unknown noise covariance and outliers. The adaptive filters assume no outliers in measurements and the robust filters assume accurate measurement noise covariance matrices. In this paper, we propose an adaptive and robust cubature Kalman filter, which achieves the adaptivity by estimating the measurement noise covariance through the variational Bayesian (VB) method, and achieves the robustness by suppressing the outliers based on the maximum correntropy criterion (MCC). The robustness and adaptivity of the proposed filter are verified through a typical tracking simulation example. Furthermore, the experimental results show that the proposed filter can obtain higher estimation accuracy than other filters in the vision/dual-IMU integrated system.