Model-based object recognition: a truth maintenance approach

A description is given of the visual constraint recognition system (VICTORS), which uses a truth maintenance system (TMS) approach to model-based vision. By incorporating an assumption-based TMS (ATMS), VICTORS can identify all occurrences of a given figure within a scene of randomly overlapping rectangles, subject to variable figure geometry, input data from multiple sources (e.g. stereo data), and incomplete figures. VICTORS thus has many advantages over existing model-based vision systems, and shows the power of using ATMs for detecting multiple figures within this domain. It also demonstrates ways in which the ATMS can be improved to increase its range of applicability