A machine-learning approach to multiple-detection data association for ASDE radar

Some types of sensors provide multiple detections per target, such as ASDE radar, infra-red or video cameras, bringing new challenges to the tracking systems based on their data. In these cases, data association needs to be extended to address blobs re-connection and target segmentation issues, losing the assumptions handled by classical approaches. In this work, the design is partially considered as a data analysis process performed over representative samples to infer appropriate rules for data association. The proposal is to apply a machine-learning paradigm based on available data samples and performance results assessed through simulation. It extends a previously proposed approach based on a efficient search in the hypotheses space, applying now data mining to develop a suitable heuristic function. The advantages of this methodology are analyzed in two representative complex scenarios of airport surface.