Matching polygon fragments

Abstract Many computer vision systems model objects using polygons. If objects occlude each other or do not appear entirely in view, we need to match a polygon scene fragment with the polygon representation of the model objects. In this paper we present a way to match polygon fragments. Using polygon moments and cross moments we can compute a dissimilarity measure between two fragments. In addition, we also find the coordinate transform that maps one fragment onto the other. These polygon moments can be conputed by using just the end points of the line segments. If the dissimilarity measure is less than a small number, then we can preliminarily conclude fragments. Otherwise the polygon fragments are dissimilar and are not considered any further. In computing the dissimilarity measure, we will also find a coordinate transform that maps one fragment to another. By using this coordinate transform, we can develop confirmation tests for determining whether the scene polygon fragments really belong to an occluded object.