Registration Error-resistant Track-to-track Association for Multiple Sensors

The process of Multi-sensor Track-to-Track Association (MTTA) involves the association of tracks observed by multiple sensors, which are of the same target. This process serves as a foundation for enhancing track continuity and refining state estimation via the fusion of multi-sensor information. Predominantly, the existing methods employ statistical modeling of target motion to devise association gates between tracks. However, these methods encounter a decrease in association accuracy when faced with unknown registration errors. To address this challenge, a data-driven approach is proposed, which reconfigures track association into a classification problem. This approach harnesses target kinematic characteristics and track trajectory features extracted at the track level. Subsequently, these features are employed in constructing a track association algorithm based on decision trees. Experiments are conducted on a publicly available dataset, Multi-source Track Association Dataset (MTAD) derived from real data. The results have demonstrated superior association performance, particularly under conditions of inadequate system registration.