A Model Decomposition Kalman Filter for Enhanced Localization of Land Vehicles
Ge Guo, Jiageng Liu, Xiaozheng Sun
Abstract
This article investigates a GPS-dead reckoning fusion localization problem for land-vehicles. We first give a nonlinear vehicle kinematics model, which is used as the transition model for the fusion filter. Then a decomposition rule is designed to resolve the transition model into a set of model families. Based on decomposed model series expansion, a fusion filter is derived to estimate the vehicle position based on Gaussian approximation. The fusion filter can guarantee higher real-time localization accuracy than other linearization-based approaches, since it is derived on the basis of the decomposed process model and the series expansion scheme involves a loss function to ensure minimum truncated orders. Numerical simulations and real-world experiments results demonstrate that our method is more accurate and reliable than the state-of-the-art methods for vehicle localization under various driving conditions.