Real-time Vision using FPGAs, GPUs and Multi-core CPUs

Computer Vision has gone through a significant evolution in the last decade, whilst finding its way into more and more application domains, with robotics, video surveillance and automation being the most dominant. This trend is due to extensive basic research in the field of vision, however the introduction and evolution of a wide variety of powerful hardware architectures have made the developed theory more applicable in performance demanding and real-time1 applications. Three different architectures have dominated the field due to their parallel capabilities that are often desired when dealing with massive amounts of visual input; the Field Programmable Gate Array (FPGA), the Graphics Processing Unit (GPU) and the multi-core Central Processing Unit (CPU). FPGAs and GPUs have been available for the past decades, but only in their most recent versions have they become interesting as general purpose proces-sors in the vision community. The introduction of programming languages like CUDA from NVIDIA has made it easier to utilize the high parallel processin