Parallel Image Understanding System on Recursive Torus Architecture

本研究的目的是开发用于并行图像理解的硬件和软件系统。1.基于递归环面结构设计了一个并行图像理解机器RTA/1(PE：1,024)，并开发了一个具有17个Transspurs的原型机器RTA/0。2.基于RTA/0的实验结果，对RTA/1的各种基本硬件性能进行了评估，证明了RTA/1.3体系结构的有效性。3.提出了一种基于RTA/1的数据级并行处理方案，并在RTA/0上进行了性能评估，证明了该方案的有效性。实验结果表明，所提出的硬件和软件系统都能达到很高的加速效果。

The aim of this research is to develop hardware and software systems for parallel image understanding. We obtained the following results by three years project.1.We designed a parallel image understanding machine RTA/1 (PE : 1,024) based on the Recursive Torus Architecture and developed a prototype machine RTA/0 with 17 Transputers.2.Various fundamental hardware performances of RTA/1 were evaluated based on experimental results using RTA/0, which demonstrated the effectiveness of the architecture of RTA/1.3.We proposed a scheme of data level parallel processing on RTA/1. Its performance was evaluated based on experiments using RTA/0, which proved utilities of the proposed scheme.4.We implemented a parallel object recognition system on RTA/0 using parallel C language. Experimental results showed that very high level speed-up can be attained by our proposed hardware and software systems.