3D object recognition using invariant feature indexing of interpretation tables

Abstract

The authors propose a technique for 3D object recognition from range data; the technique collects partial interpretations from precompiled tables indexed by invariant features of pairs of surface patches. This technique was motivated by some recent work in so-called geometric hashing procedures. A technique for automated construction of invariant feature indexed interpretation tables from CAD models is described. Recognition using these tables consists of an interpretation creation phase and an interpretation extension phase. The authors describe both procedures, and test them on some real images. The use of precompiled tables of hypotheses offers the potential for greater efficiency in the recognition step than techniques such as constrained search. >