The track repulsion effect in automatic tracking

The track coalescence effect degrades the performance of probabilistic data association trackers in dense target scenarios. Recently, it has been observed that an opposite effect exists with trackers that utilize hard data association, which we denote as the track repulsion effect. In this paper, we examine this effect in the context of a crossing target scenario, and explore the effectiveness of a track-oriented multi-hypothesis tracker in combating this effect, with both single-stage and multi-stage processing configurations.