An adaptive particle filter for MEMS based SINS nonlinear initial alignment

The MEMS based SINS initial alignment with large azimuth is a nonlinear and non-Gaussian filtering problem. The particle filter (PF), is a popular estimation method for such problems. In order to realize initial alignment for MEMS based SINS combined with magnetic compass, a particle filterer method which uses an Extended Kalman Filter (EKF) to generate the mean and covariance of the importance proposal distribution is developed. In order to reduce the computational burden, an adaptive extended PF (AEPF) is proposed. The relation between the filtering accuracy and the sampling number drawn by Particle Filtering based on the confidence interval theory is introduced. We adjust the number of particles according to the filtering precision. Simulation results demonstrate that the new adaptive particle filtering method can obtain a better performance compared with the conventional PF with the reduction of computational load.