Fusion Estimation for Nonlinear Systems with Heavy-tailed Noises

Abstract

In some target tracking scenarios, the process noise and the measurement noise are both heavy-tailed noises. This type of noise can not be modeled as Gaussian noise, since it has quite different characteristics. Existing algorithms for fusion estimation of nonlinear systems with Gaussian noises are no longer applicable for systems with heavy-tailed noises. In this paper, estimation of multisensor data fusion for nonlinear systems with heavy-tailed process noise and measurement noise in target tracking is studied. Based on the robust Student’s t based nonlinear filter (RSTNF), a filtering method using unscented transformation (UT) for state estimation of nonlinear systems with heavy-tailed noises, we present a modified nonlinear filter in case of package drop out exists. For fusion estimation of multisensor nonlinear systems, we present the centralized fusion based on the modified filter. Our results from Monte Carlo simulations on a target tracking example demonstrate the effectiveness and the robustness of the presented algorithm.