Multiple Hypothesis Tracking for Processing Tracklets

Abstract

Multiple hypothesis tracking (MHT) addresses difficult data association problems by maintaining multiple association hypotheses over multiple frames of data. In many multiple target tracking (MTT) problems, sensor measurements that originate from the same targets can be grouped into tracklets for further processing by another tracker. This processing approach improves the efficiency of MHT because association is performed on tracklets instead of measurements. This paper introduces two types of tracklets: pure tracklets representing single targets and ambiguous tracklets representing multiple targets. It discusses how MHT can be used for stitching tracklets from a single sensor and associating tracklets from multiple sensors.