Multiple Person Tracking in Advanced Meeting Environments
S. Schreiber, Gerhard Rigoll · 2005
Abstract. This paper addresses the problem of tracking an unknown number of persons in meeting scenarios with a monocular camera. Although there might arise several problems caused by cluttered or noisy video data, it will be shown in this proposal that tracking of multiple persons in such scenarios can be realized by our novel idea. Our architecture basically utilizes a factored sampling technique- also known as ICondensation proposed by Isard and Blake [2]- to generate several hypotheses of a possible location for the tracked object. To enable a stable tracking we implemented an approach combining flexible properties of active shape models (introduced by Cootes et al. [1]) with the particle filter framework mentioned above and thus we could achieve robust results for tracking human heads in cluttered environments. Usually hypotheses will concentrate over time on the most likely location for the head in the image, thus a hierarchical structure has been invented which indicates both the number of persons and their position.