Robot vision based on three‐dimensional model

Abstract

Abstract It is desirable for flexible robot vision to have three‐dimensional models have no constraints are imposed on the scene. This paper proposes the use of geometric models generated during the CAD process to save model building efforts. Also, a hierarchical matching method to avoid search efforts for all possible attitudes of each model is proposed. Examples are described for three cases: 1) only a few surface normals are available; 2) all visible surface normals are available; and 3) range data are available. In all cases, efficient and flexible object recognition is realized by hierarchical matching based on three‐dimensional models.