Cina Motamed
Laboratoire LASL, Universite du Lottoral Cote d'Opale, Calais, France.
A video-surveillance system for human and car activity monitoring is presented and evaluated over a car park test sequence. Our strategy is to illustrate how history of events can help decisions in the video-surveillance algorithms. In the algorithms this history reflecting the persistance of a fact is resumed as a indicator updated at each sequence. = =20 The first stage of the video surveillance concerns the motion detection process in order to highlight interesting object. An updating of the back ground reference is proposed in order to control changes in illumination and the object integration. The control of this object integration at each pixel is based on the stabibility indicator of the current reference . The tracking step is a 2D approach of region tracking working over a variety of applications. An associate belief indicator resuming history of tracking quality for each object permits to manage a majorities of ambiguities. These algorithms are tested with pets-2000 data-set, and discussed particularly when ambiguities appear.
References
[1] P.VANNOORENBERGHE, C. MOTAMED, J.-M. BLOSSEVILLE & J.-G. POSTAIRE."A motion detection for non-rigid objects. Application to pedestrians monitoring in an urban environment" IEEE International Conference on IMACS, Computationnal Engineering in Systems Applications, Lille, France, 9-12 July 1996, Volume 4, pp 438-442. [
[2] C.MOTAMED. 'Motion detection and tracking for human activity monitoring' 2nd Conference on Information fusion, Fusion'99, San Francisco , USA July 6 - 8, 1999.
[3] VANNOORENBERGHE, C. MOTAMED & J.-G. POSTAIRE " Updating a backgroud image for motion detection in urban area " (in french) Revue Traitement du signal, 1998 Vol 15 n2 p139-148.
[4] C.Motamed, P.Vannoorenberghe. ' Behavioural knowledge for video surveillance''proc. Of Workshop ''Dynamic scene recognition from sensor data'' Toulouse, Juin 1997.
[5] P.VANNOORENBERGHE, C. MOTAMED, J.-M. BLOSSEVILLE & J.-G. POSTAIRE. 'Automatic Pedestrian Recognition using Real Time Motion Analysis' Conference : ICIAP 97 Image Analysis and Processing, Florence, Italia, Septembre 1997, pp 493-500.