A distributed artificial intelligence view on general purpose vision systems (abstract)

Abstract

A General Purpose Vision System (GPVS) is an open and domain-independent system that is able to build, maintain, and use an internal representation of the external world from data provided by physical sensors such as cameras. The experience in constructing the Vision As Process (VAP) and SATURNE systems using a Distributed Artificial Intelligence (DAI) approach has enabled us to compare, from a DAI point of view, these two systems with other major existing GPVS's. As far as we know this work represents the first large-scale comparison between GPVS's at the module and module-integration levels using a DAI formalism. This has lead us to the identification of several basic common features within the systems studied. Furthermore we think these common features are essential for the construction of a GPVS, namely: 1/the distribution of the knowledge representation provides level-agents that denote the same knowledge at a given level of representation 2/the distribution of the knowledge processing provides focus-agents which realize the system's "focus of attention" 3/intersection of level-agents and focus-agents determines the active basic-agents that explicitly or implicitly constitute the system at a given time. The description of GPVSs in terms of basic-agents is a novel feature of our approach and may also be applicable to other domains such as robotics.