A 3D model based visual surveillance system

This paper presents a visual surveillance system tailored for remote surveillance of indoor environments. It uses a two layers video codec based on a 3D model of the background [2]. The objective is to achieve high data compression rates and to enable the support for applications that require the 3D knowledge of the scene. Examples of such applications are statistical applications (e.g. counting persons) and security applications (e.g. control of entrances in restricted areas). The system supports the ability to localize moving objects in the 3D space assuming that all objects are placed on the ground plane. This feature allows the localization of people on a top view map of the site. Results of some tests are presented in order to show the system ability to deal efficiently with the motion of the camera and to illustrate the two layers separation based on a background/foreground segmentation.