Spatiotemporal multiple persons tracking using Dynamic Vision Sensor

• Published in: 2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops pp. 35 - 40
• Main Authors: Piatkowska, E., Belbachir, A. N., Schraml, S., Gelautz, M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2012
• Subjects: Clustering algorithms; Data models; Dynamics; Heuristic algorithms; Machine vision; Tracking; Voltage control
• DOI: 10.1109/CVPRW.2012.6238892
• ISBN: 1467316113; 9781467316118
• ISSN: 2160-7508

Although motion analysis has been extensively investigated in the literature and a wide variety of tracking algorithms have been proposed, the problem of tracking objects using the Dynamic Vision Sensor requires a slightly different approach. Dynamic Vision Sensors are biologically inspired vision systems that asynchronously generate events upon relative light intensity changes. Unlike conventional vision systems, the output of such sensor is not an image (frame) but an address events stream. Therefore, most of the conventional tracking algorithms are not appropriate for the DVS data processing. In this paper, we introduce algorithm for spatiotemporal tracking that is suitable for Dynamic Vision Sensor. In particular, we address the problem of multiple persons tracking in the occurrence of high occlusions. We investigate the possibility to apply Gaussian Mixture Models for detection, description and tracking objects. Preliminary results prove that our approach can successfully track people even when their trajectories are intersecting.