Concluding Remarks

Abstract

This book has addressed several issues in multiprocessor architectures and parallel algorithms for Integrated Vision Systems. The approach has been to consider computational requirements for vision applications in an integrated environment in designing a multiprocessor architecture rather than to propose architecture solutions to perform individual algorithms efficiently. An IVS involves algorithms from several levels of processing and the characteristics of algorithms in each level differ tremendously from algorithms in other levels. However, these algorithms need to exist in a system simultaneously and interact with each other. Therefore, a multiprocessor architecture suitable for IVS applications must be partitionable and reconfigurable. It must have the capability to allocate resources dynamically, provide for dynamic load balancing and task scheduling, provide fast and flexible communication, provide efficient I/O, and be fault-tolerant, in addition to providing raw processing power.