Architectural considerations for highly scalable computing to support on-demand video analytics

The processing demands on video analytics calls for special design considerations to achieve scalability. Numerous factors influence the running time of an analytics job. The time consumed for raw computing can be improved by well-engineered approaches to execute certain sub-tasks. High scalability can be achieved by selectively distributing computational components. We elucidate such factors that aid scalability and present design choices for architecting them. The principles outlined in this research were used to implement a distributed on-demand video analytics system that was prototyped for the use of forensics investigators in law enforcement. The system was tested in the wild using video files as well as a commercial Video Management System supporting more than 100 surveillance cameras as video sources. The architectural considerations of this system are presented. Issues to be reckoned with in implementing a scalable distributed on-demand video analytics system are highlighted.