Analysis control middleware for large-scale video surveillance

Recently the demand for large-scale surveillance systems automated by video analysis technologies has been increased. A key to developing real-world systems of this type is the streamlining of analysis execution on the basis of application characteristics including video contents. This paper proposes novel middleware we call “Analysis Control Middleware (ASCOT)” for achieving and customizing analysis execution control. It provides an Analysis Control Framework for load reduction, load control and customization of control logics in addition to stream processing features. We applied ASCOT to a case study application consisting of a search function for forensics and a suspect alert function for real-time intruder detection. For a system with hundreds camera, ASCOT reduces the number of servers by 40.2% by combining two simple analysis controls. Furthermore, the effects on load control and customization of control logics on the middleware are evaluated with the case study application.