Towards the optimal hardware architecture for Computer Vision

This chapter addresses an analysis of different computing paradigms and platforms oriented to image processing. Previously, a representative set of Computer Vision algorithms covering the three levels of processing is reviewed. This study will lead us to observe the algorithms in terms of a set of common characteristics: operations, data type, program flow, etc. This is critical to design new hardware architectures in order to maximize performance. The analysis from the hardware point of view will highlight the best features of the most used computing paradigms in order to establish a relationship between the type of operation, data, programming model and hardware architecture. An efficient architecture for Computer Vision must combine all the selected features. The analysis of the characteristics of the different algorithms will lead us naturally to an optimized general-purpose hardware architecture for Computer Vision.

keywords: Machine Vision, Computer Vision, Hardware Architectures, FPGA, DSP, Massively-Parallel SIMD chips, Vision Chips.