Versatile real-time vision based on a distributed system of personal computers

Abstract

This paper reports the development of a cost-effective framework for the implementation of a versatile computer vision system, namely the Cyvis-1. Presenting a strong biological motivation, the Cyvis-1 approach is characterized by the multi-stage integration paradigm, which is inherently suitable for parallelization and representation through the OOP/GUI concepts. A model for the parallelization of this system has been developed that incorporates multitasking, distributed processors, and message interchange through channels. The implementation of each of these parallel features in IBM-PC compatible machines running under Windows 95/NT has been described. An application example of RGB image edge and line detection has been presented and assessed in terms of execution time, efficiency of hardware utilization and speed-up.