A Gaussian Distributed Resampling Algorithm for Mitigation of Sample Impoverishment in Particle Filters

Abstract

This paper proposes a new resampling algorithm, a Gaussian distributed resampling (GDR) algorithm, in order to solve the sample impoverishment problem in particle filters. The key concept of the proposed algorithm is to generate new particles on the basis of a Gaussian distribution, which depends on the size of the weights in the resampling process. In comparison with established resampling algorithms, particle diversity can be maintained, and thus the proposed algorithm avoids sample impoverishment. The proposed GDR algorithm guarantees a reliable estimation even if the number of samples is sharply reduced. Thus, the computational burden of particle filters can be reduced efficiently with the proposed GDR algorithm.