Gaussian Mixture Filter Based on Variational Bayesian Learning Optimization and Its Application to Integrated Navigation

We present a novel Gaussian mixture filter (GMF) improved by variational Bayesian learning. The method is mainly used in time-differenced carrier phase/strap-down Inertial Navigation System integrated navigation to deal with parameter estimation of Gaussian mixture model in non-Gaussian noise environment. By means of variational Bayesian learning theory, this algorithm accurately and efficiently estimates parameters of GMF, and further refines the stochastic model. And it improves accuracy of GMF, reduces the computational complexity, and improves the effectiveness. The experimental results illustrate that the proposed filter substantially outperforms the existing algorithm in terms of estimation accuracy, and is computationally much more efficient. It provides theoretical support for integrated navigation data fusion strategy based on GMF.