Detection of intrusion using a background structural model

In this paper, the problem of intrusion diagnostic in uncontrolled environment, that means without any restrictions concerning the scene features, is discussed. We propose an automated surveillance system based on dynamic image sequences processing in order to detect intrusions in a specified surveillance area. An original method of the background representation using structural features contributes to detect the changes of the scene due to the intrusions of the moving objects. The described algorithm models the region of interest by the computation of a gradient mask with coefficients depending on the background structure. As gradient images are not sensitive to light variations, the experimentation shows that the proposed algorithm allows to take into account the illumination component changes and thus minimises the false alarms rate. The performance of this new intrusion detector are illustrated on an image sequence under several illumination conditions.