Applied technology in adapting the number of particles while maintaining the diversity in the particle filter

Determining the required number of particles is a challenging task toward the application of particle filters (PF). As smaller number of particles means faster computing speed while larger number of particles means better approximation ability, it is vital to balance the trade-off between computing speed and approximation quality. Moreover, to match the system requirement that often varies in time, the number of particles should be adapted in time. To achieve these, an Adaptive Deterministic Resampling (ADR) is proposed in this paper. The new resampling employs techniques combine Deterministic resampling and Kullback-Leibler Divergence (KLD)-resampling (both with slight changes made) and gain their respective advantages, which not only allow online adaptation of the number of particles according to the system requirement but also guarantee the diversity of particles. Simulation results demonstrate and confirm the validity of our approach.