2008 Reports

Exploring Computation and Communication Trade-offs in the Design of Automatic Video Surveillance Networks

Pinto, Alessandro; Carloni, Luca; Sangiovanni-Vincentelli, Alberto

Video surveillance is one of the fastest-growing class of networked embedded systems. An increasing number of cameras are networked to support various applications including security in city streets, emergency evacuation in large buildings and direct marketing in department stores. The large number of networked cameras motivates the need for automatic video analysis, which, as of today, relies mostly on centralized computation. Still, trends in embedded computing enable the cost-effective realization of smart camera nodes and, consequently, the distribution of part or all of the computation. Starting from a particular application of automatic video surveillance for building automation, we derive a system-level model of the main computational tasks that are necessary to process a collection of video streams together with their requirements in terms of computation and communication resources. Then, we define a set of alternative implementation platforms based on a detailed analysis of the possible choices in terms of off-the-shelf components and interconnection network technologies. Finally, we present a methodology and supporting CAD tool that assists us in evaluating alternative partitioning/mapping of the computational tasks onto the various platforms.