Intelligent Imaging System

Intelligent imaging systems of the future require Teraops of arithmetic performance with low power dissipation. However, they also demand full programmability and functionality over a wide range of application characteristics present in intelligent imaging systems. Stream processors have recently emerged as a technology that can provide Teraops of performance at a power effieincy of 200 GOPS/W of performance with full programmability from high-level C. This technology has broad application in surveillance, UAVs, mobile robots, and many other computer vision and intelligent image processing systems. We propose to evaluate stream processors as a single-chip solution for meeting the processing and programmability requirements in these imaging systems. A stream processor based hardware platform will be developed, imaging processing performance on key applications will be evaluated, and comparisons to competitive technologies such as the hybrid CNN-UM/DSP approach will be carried out. Furthermore, advanced "intelligent" imaging functions such as quasi-connected component, target motion analysis, and location/appearance predictions will be mapped to the stream programming model in order to evaluate their performance on stream processors.