Relational models for object location

Abstract

The introduction of vision into advanced automation technology has been impeded by our lack of understanding of the cognitive processes involved. Significant theoretical advances towards explicating the "ostensive" nature of perceptual events has yet to emerge. We have, correspondingly, chosen to investigate empirically a subset of the general problem of machine vision. In particular, we are interested in the rapid and reliable recognition of the location of a known object. As a first step in this direction, we have identified the important role of "relations" in descriptive schemata, as opposed to "properties", in capturing an essential connotation of ostensive description. New combinatorial problems associated with relational descriptions became the focus of the research effort.