One adaptive centralized Kalman filter for multisensor combined navigation system based on variational Bayesian method

Abstract

The filtering method of multi-sensor combined navigation systems is usually based on the centralized Kalman filter(CKF), and assumes that the measurement noise covariance matrix (MNCM) is known. However, in actual situations, the MNCM is unknown or changes over time. Therefore, this paper first establishes an adaptive filtering model for a single combined navigation system based on the Variational Bayesian (VB) algorithm to estimate the changes in the MNCM more accurately. Then, an adaptive centralized Kalman filter (ACKF) based on the VB method (VB-ACKF) is established, based on the system model of the multi-sensor combined navigation system. Finally, the above algorithms are verified using the GNSS/CNS/ADS/SINS multi-sensor combined navigation system as an example. The experiment demonstrates that VB-ACKF can accurately estimate the changes in MNCM, and VB-ACKF has better filtering accuracy than CKF.