Real-time tabu search for video tracking association

Abstract

Intelligent Visual Surveillance (IVS) systems are becoming a ubiquitous security component as they aim at monitoring, in real time, persistent and transcient activities in specific environments. This paper considers the data association problem arising in IVS systems, which consists in assigning blobs (connected sets of pixels) to tracks (objects being monitored) in order to minimize the distance of the resulting scene to its prediction (which may be obtained with a Kalman filter). It proposes a tabu-search algorithm for this multi-assignment problem that can process more than 11 frames per seconds on standard IVS benchmarks, thus significantly outperforming the state of the art. © 2009 Springer Berlin Heidelberg.