Nathanael Rota, Robert Stahr and Monique Thonnat
Orion Group, INRIA Sophia Antipolis, Nice, France
In this article we present the tracking algorithm used in the Visual Surveillance system VSIS (Visual Surveillance Intelligence System) developed in the Orion group at INRIA Sophia Antipolis. The research area of the Orion group is not tracking but AI. Our goal is to recognize the behaviour of the object (person) being tracked. The algorithm we present here is therefore far from entirely our invention. The algorithm consists of four phases: motion detection, clustering, merging and suppression of clusters and frame-to-frame matching of clusters. The innovation is the use of 3D contextual information in the merging/suppression and matching phases. The VSIS tracker works on monocular (colour or B/W) image sequences from a single immobile camera with fixed focal length and is currently optimized for person tracking.