A structured mean field approach for existence-based multiple target tracking

Abstract

A challenge of multiple target tracking is that targets appear and disappear without notice. To overcome this difficulty, we propose a joint estimation algorithm, Structured Mean Field Existence-based Tracker (SMFET), that infers not only the states of targets but also the existence of targets. SMFET is related to Joint Integrated Probabilistic Data Association (JIPDA) through their existence model. Although JIPDA is effective in the presence of false alarms and missed detections, the algorithm suffers from track coalescence. Unlike JIPDA, our algorithm avoids track coalescence, which we demonstrate through simulations containing a varying number of closely spaced targets. Results show SMFET to be tractable with accuracy that rivals JIPDA.