A comparison of nonlinear filtering approaches with an application to ground target tracking

With an application to ground target tracking, two groups of nonlinear filtering approaches are compared in this paper: Gaussian approximation and Monte Carlo simulation. The former group, consisting of the extended Kalman filter (EKF), Gauss-Hermite filter (GHF) and unscented Kalman filter (UKF), approximates probability densities of nonlinear systems using either single or multiple points in a state space, while the latter group, being particle filters, estimates probability densities using random samples. There are two sources contributing to nonlinearity in the ground target tracking problem: terrain and road constrained kinematic modeling and polar coordinate sensing. When tracking ground maneuvering targets with multiple models, one faces another problem, i.e., non-Gaussianity. This paper also compares interacting multiple model (IMM)-based filters IMM-EKF, IMM-GHF and IMM-UKF with particle-based multiple model filters for their capability in handling the non-Gaussian problem. Simulation results show that: (1) all the filters achieve a comparable performance when tracking non-maneuvering ground targets; (2) particle-based multiple model filters are superior to IMM-based filters in maneuvering ground target tracking.