The ensemble unscented particle filter
Xi Qing Wei, Xiu Jie Zhang, Han Yu, Shen Min Song · 2011
The particle filter is a Monte Carlo method that allows us to treat any probability distribution, nonlinear and non-Gaussian. However the choice of the proposal distribution is the most critical problem. Unscented particle filter (UPF) uses UKF to generate and propagate the Gaussian distribution which provides a better approximation to the optimal conditional proposal distribution. It is not practical to fulfill the requirement for large-scale problems that the number of the sigma points will be larger than twice the degree-of-freedom of the system model. To overcome this difficulty, a new particle filter equipped with ensemble unscented Kalman filter (EnUKF) is proposed with the name EnUPF. The analyses indicate that EnUPF needs less computational cost and simulation validate the similar performance to UPF.