An architecture for automated scene understanding

This paper presents distributed, automated, scene surveillance architecture. Object detection and tracking is performed by a set of region and object agents. The area under surveillance is divided in several sub-areas. One camera is assigned to each sub-area. A region agent is responsible for monitoring a given sub-area. Background subtraction is first performed on the scene taken by the camera. Based on the foreground mask, the region agent segments the incoming frame and creates object agents dedicated to tracking detected objects. Tracking information and segments are sent to a scene processing unit that analyzed this information and determined if a threat pattern is present at the scene and performed appropriate action.