The Ornstein-Uhlenbeck Process in Multi-Target Tracking
Stefano P. Coraluppi, Craig A. Carthel, Jordan LeNoach, Brandon Bale · 2021
This paper introduces a multiple-model Ornstein-Uhlenbeck (MM-OU) formulation of target motion uncertainty. This model exploits a weighted set of learned motion trajectories for targets. Each realization of target motion is governed by one trajectory in the set, with deviations according to a stable 2ndorder Ornstein-Uhlenbeck process. Based on this target model, we develop a (non-interacting) MM-OU filter that maintains a weighted set of filter solutions. This filter is useful in setting with nontrivial sensor coverage gaps. We have integrated the MM-OU filter as part of an advanced multiple-hypothesis tracking solution to the multi-target tracking problem. We examine the benefits of the new filter compared to baseline processing that adopts the classical nearly-constant velocity filter, with both single-target and multi-target scenarios.