Recognition of occluded objects using curvature

Abstract

New approaches of object representation reliable for partially occluded objects recognition are introduced in this article. Objects are represented by their boundaries, which are deformed by the occlusion. The boundary representation was made by approximation with circle arcs. The representation was designed to be local and robust to occlusion. The curve approximation with circle arcs is equivalent to the curvature representation with respect to noise. The algorithm is simple and easy to implement. Experimental results are presented.