Optimization of combined filter for vehicle navigation

Abstract

Accuracy of single filter used in most of vehicle navigation decreases when system state is both linear and nonlinear. According to the output state of the receiver to choose linear or nonlinear filter, a kind of combined filtering structure is proposed. Furthermore, a derivative-free filter is adopted as nonlinear filter, especially, formulas of state and random noise covariance matrix's Cholesky decomposition update are improved, which guarantees positive definiteness of covariance matrix and its square root; Due to high precision and robust performance of linear Kalman filter on linear filtering, it's filtering equations are improved to eliminate positioning error caused by velocity error; The advantages of two filters are verified in static and dynamic process respectively by simulation experiment. Finally, vehicle simulation results show that the combined linear-nonlinear filter has low error and computation cost, is better than single filter.