Programmable SoC processor for video object recognition and tracking applications

This paper describes a novel system-on-chip architecture for video analysis. It is shown how input video characteristics and processing functionality and speed, impact the derivation of the SoC architecture. In particular, the real-time processing demands create the need for specialized cores and optimizations applied to the memory organization to reach the performance required for VGA and HDTV video analysis. The SoC performance targets for a template matching based object recognition and tracking algorithm are presented.