Modified unscented particle filter using variance reduction factor

Sequential Monte Carlo based estimators, also known as particle filters (PF), have been widely used in nonlinear and non-Gaussian estimation problems. However, efficient distribution of the limited number of random samples remains a critical issue in design of the sequential Monte Carlo based estimation algorithms. In this work, we derive a modified unscented particle filter based on variance reduction factor that obtains an efficient distribution of the random samples using a scaled unscented transform. The proposed algorithm is shown to combine the robustness of the unscented particle filter with relatively low computational complexity of the generic particle filter. The efficiency of the proposed approach is evaluated in nonlinear problem of bearings-only target tracking, and its performance is compared to the regularized PF and the Cramer-Rao low bound.