Abstract

A new method of locating partially visible two-dimensional objects is presented. The method is used to locate complex industrial parts that may contain several occurrences of local features, such as holes and corners. The matching process utilizes clusters of mutually consistent features to hypothesize objects and also uses templates of objects to verify these hypotheses. The technique is fast because it concentrates on key features that are automatically se-lected on the basis of a detailed analysis of computer-aided design (CAD) models of the objects. The automatic analysis applies general-purpose routines for building and analyzing representations of clusters of local features that could be used in procedures to select features for other lo-cational strategies. These routines include algorithms for computing the rotational and mirror symmetries of objects in terms of their local features. Key terms: object recognition, two-dimensional part lo-cation, industrial vision, symmetry, computer vision, plan-ning.