Particle filter Guided by Iterated Extended Kalman filter

In particle filter (PF), the resampling step effectively solves the problem of particle degeneracy. However, it introduces the new problem of particle impoverishment. To tackle this problem, a PF Guided by the Iterated Extended Kalman filter (IEGPF) is proposed. Firstly, a maximum likelihood ratio (MLR) is defined to measure how well the particles, drawn from the transition prior density, match the likelihood model. Then, according to the MLR, particles are adaptively divided into two groups. Those in one group are drawn from the transition prior density, while those in the other group are drawn from the Gaussian approximate posterior density, obtained by the iterated extended Kalman filter (IEKF). Compared with traditional sampling strategies, the proposed strategy is more flexible for time-varying system characteristics, e.g., measurement noise variance. Simulation results demonstrate the improved performance of IEGPF over the Sampling Importance Resampling (SIR) PF, the Extended Kalman PF (EPF) and the Unscented Kalman PF (UPF), etc.