Rule-driven processing and recognition from range images

Abstract

A rule-driven recognition system of a 3-D scene using range images is described. The system is constructed with a production system. A scene is described with simple a surface patch network (planar or quadric) by examining the local curvature of objects. The region and edge network is interpreted with IF-THEN-like heuristic rules. The system can recognize smoothly joined curved surfaces and explain the structure of a scene consisting of plural objects. The experiments have been carried out with real range data.<>