Adaptive Particle Filtering with Variational Bayesian and Its Application for INS/GPS Integrated Navigation

This paper considers the unknown measurement noise covariance problem in the nonlinear situation of a navigation system. Aiming at the contaminated GPS signals and the outlier environment, there are many variational Bayesian based Gaussian approximation methods in the integrated navigation system. However, the integrated navigation is nonlinear, especially for the inaccurate initial state that provided by the initial alignment stage. The variational Bayesian method is firstly incorporated into cubature particle filter, of which the proposal distribution is set as cubature Kalman filter to provide the accuracy and stable estimation for the the application of navigation system. Then KLD resampling method is merged into the variational Bayesian based cubature particle filter to supply the sufficient particles that enhance the stability of the proposed filter. The numerical simulation demonstrates that the proposed filter do better at presenting accurate and stable estimation than the variational Bayesian based cubature Kalman filter, and the experiments verify the effectiveness of the proposed filter for the application of integrated navigation system.