TF-TS: Track-to-Track association via track fusion and track segmentation in distributed multi-sensor systems

Abstract

Track-to-track association (T2TA) aims at unifying batch numbers of tracks, reducing redundancy, and clarifying the situation. It is the precondition and foundation of track fusion, situation awareness, and traffic control. Existing methods based on statistical reasoning or fuzzy math bring some problems that are difficult to solve simultaneously, such as unreasonable assumptions, unsuitable models, uncertain thresholds, and long association times. In the light of the above problems, in this paper, we proposed a T2TA method via Track Fusion and Track Segmentation (TF-TS). The track fusion module fuses and extracts track features from several tracks to reduce the dependency on prior assumptions, motion models, and thresholds. The track segmentation mapping module transforms track tensors into association matrices directly to improve association efficiency. With four kinds of constraints, the association matrices inferred by TF-TS are close to the real one. TF-TS can reduce the dependence on the assumptions, motion models, and thresholds. It can also reduce the traversal calculation of tracks and increase the association efficiency. The global AIS tracks are used to train and test our model, and association results demonstrate that the proposed method can associate tracks in complex scenarios. Moreover, the efficiency is improved and the demand for the real-time association is satisfied