New Box Particle Filter with Improved Resampling Method and Extended Inclusion Volume Criteria for Multi-target Tracking

Abstract

In the resampling procedure of traditional box particle filtering, selected box particles are divided in a randomly chosen dimension. This resampling procedure may fail when some elements in the target state vector are unmeasured. To deal with this problem, an improved resampling method for box particle filtering is proposed, where a limit on the sizes of box particles is imposed to restrain the box particles from growing too large. In addition, we extend the inclusion and volume criteria from single-target tracking to multi-target tracking. Instead of indicating whether the true target state is included in the support of the posterior track probability in single target tracking, the inclusion value in multi-target tracking indicates how many true targets are included in the supports of the posterior probability densities. And the volume value in multi-target tracking is redefined as the mean volume of the supports of the posterior probability densities. Simulation results are provided to illustrate the effectiveness of the proposed approach.

Keywords

• Box particle filter
• multi-target tracking
• resampling
• inclusion and volume
• labeled multi-Bernoulli filter