Smoothing for Nonlinear Multi-Target Filters with Gaussian Mixture Approximations

摘要

This paper investigates a smoothing method using the nonlinear Gaussian mixture probability hypothesis density (GMPHD) filter for use in multi-target tracking. This specific smoother is developed using backwards recursion operations in order to improve upon the preexisting forward filtering solution. The observational and dynamical models considered are nonlinear in nature, creating complexities not present in previous works that developed multi-target smoothers for linear dynamics and measurements. The nonlinear GMPHD smoothing solution is compared to established smoothing solutions to test the validity of the derived algorithms, and Gaussian mixture splitting is implemented to help address common operational problems experienced by the smoother.