Application of AUV Navigation Based on SVD Unscented Particle Filter

In order to solve the problems of low filtering accuracy and poor real-time performance of standard algorithm in strong nonlinearity, a SVD Unscented Particle Filter (SVDUPF) based on strong tracking and Kullback-Lerbler Distance (KLD) resampling is proposed. Firstly, the stability of covariance matrix is guaranteed by singular value decomposition. Then the multi-fading factor adaptive adjustment covariance matrix in the strong tracking theory is introduced. Finally, the number of particles required for the next filtering iteration is determined according to the KLD sampling principle. Eliminating unnecessary particles and using the minimum number of particles on the premise of ensuring accuracy can improve navigation accuracy, achieve powerful tracking and reduce computational complexity. By comparing the results of the simulation Extended Kalman Filter (EKF) algorithm, the feasibility and application of the Unscented Particle Filter (UPF) algorithm and SVDUPF in AUV navigation positioning algorithm are tested. The correctness of SVDUPF algorithm is proved.