A Distributed Systems Architecture for Real-Time Computer Vision

This paper describes a parallel, distributed systems architecture for real-time computer vision. The systems architecture is based on a collection of autonomous agents, which run concurrently, communicating through a fixed network topology in order to achieve the system's goals of visual processing. By distributing the processing across several processors, greater performance can be achieved than on any single processor. The overall visual processing strategy is based on the concepts of: pipelined parallel processing, multi-resolution processing, foveation (focus of attention), and prediction/verification. The architecture has been used to implement several real-time tracking systems, most important of which are the Feature-Based Tracker (FBT) and its 3D version (3D-FBT). 1 Real-Time Task-Directed Robot Vision Most traditional computer vision has what might be called a "data-processing" mentality. Under this view, the goal of a computer vision system is to produce, as output, an inter..