Adaptive unscented particle filter with KLD-Sampling for nonlinear state estimation

The Unscented Particle Filter (UPF) was considered as one of the most effective state estimation method for nonlinear and non-Gaussian system. However, UPF had the inherent drawback of costly calculation. An Adaptive UPF by online choosing the number of particles was proposed to overcome the drawback of computational burden in the traditional UPF. The KLD-Sampling was used to determine the number of particles of adaptive UPF. The new algorithm chose a small number of particles if the density was focused on a small subspace of the state space, and it chose a large number of samples if the state uncertainty was high. The computer simulations were performed to compare the Adaptive UPF algorithm and the traditional UPF in performance. The simulation results demonstrate that the Adaptive UPF is very efficient and smaller time consumption compared to traditional UPF. Therefore the Adaptive UPF is more suitable to the nonlinear and non-Gaussian state estimation.