Extended GM-PHD filter for multitarget tracking in nonlinear/non-Gaussian system

The Gaussian mixture probability hypothesis density (GM-PHD) filter involves the joint estimation of number of targets as well as their individual states in linear/nonlinear Gaussian system, however theoretically not suit for dynamics with non-Gaussian noise. In this paper we show that for the state transition and likelihood both with non-Gaussian distribution, the prior and posterior PHD still can be formulated by the weighted Gaussian sum and propagating the Gaussian mixtures separately over time, in this way, state estimation in a nonlinear/non-Gaussian system can be approximately recast as state estimation in a set of parallel nonlinear/Gaussian systems, moreover, the reduced rank scaled unscented/ensemble transform variational (RSEV) filtering [11] is applied to each individual nonlinear/Gaussian system for an improved accuracy of estimation of Gaussian pdf. In addition, an implementation of the proposed algorithm is proposed by combining the closed-form recursions with a strategy for pruning/merging to the number of Gaussian components to increase efficiency.