Supporting Real-Time Active Visual Surveillance in Wireless Network*

In this paper, we study the design of an efficient wireless visual surveillance system. Two important considerations are to minimize: (1) the video workload on the wireless network; and (2) the processing workload at the front-end video capturing units. To achieve the first objective, we propose a cooperative framework for semantic filtering of video frames instead of forwarding every video frame to the backend server for analysis. To minimize the processing workload at the front-end unit, a hierarchical object model (HOM) is designed to model the status of the objects in a video scene and their temporal and spatial properties in the scene. With the information provided from the backend server, the front-end unit analyses the current status of the objects in the HOM by comparing with the selection conditions in the submitted monitoring queries. An adaptive object-based evaluation (APOBE) scheme is proposed to reduce the processing workload at the front-end unit in which a higher evaluation frequency is given to the object which is closer to satisfy the condition in the monitoring queries.