Construction of a Parallel Vision System that Satisfies High Speed, Versatility and Scalability

点击翻译按钮获取中文摘要

Aim of this research is to realize such a novel parallel vision system that satisfies requirements of high speed, versatility, and scalability. Conventionally these requirements are incompatible each other. The authors have overcome the problem of incompatibility by augmenting Beowulf-class PC cluster with dedicated hardware for visual processing. Key idea is introduction of video bus. Not only a link for inter-processor communication, such as ethernet, but also a link for video data transfer(video bus)is introduced to the system. Because of the video bus, input video image is fed to each element PC of the cluster simultaneously without any time-wasting duplication of the input image. This feature guarantees continuous and high speed(video rate)processing. Versatility and scalability are obtained by the nature of Beowulf class PC cluster which consists of commodity hardware.At the first stage of the three years research plan the authors have implemented a prototype system that is based upon the proposed idea described above. Then we have implemented a random accessible motion video recorder, named Virtual Video Tape(VVT), on the prototype. VVT is realized as memory based and software only. At the last stage of the research plan, we have developed some application programs using the prototype and VVT, and conducted some experiments to confirm the feasibility and effectiveness of the proposed idea. Topics of those experiments are face detection, face identification, adaptive(or environmentally tunable)discriminant analysis, and so forth. Experimental results support t possibility, feasibility and effectiveness of the proposed idea.