POINT CORRESPONDENCE CORRECTION BASED ON SURFACE CURVATURE FEATURES

Abstract

3D computer model of a real object has been widely used in various applications such as motion capture, computer vision and computer graphics. To build a 3D computer model, multiview data point cloud of real object from different view point that obtained from a 3D scanner must be registered to placing the multiview data point cloud into  a common coordinate system. Correspondence to find pair point matching is an important step in registration. False correspondence will affected to the registration quality.A novel technique of point correspondence correction between two point clouds is presented in this paper. The correspondence technique is started by selecting pair point matching candidate base on two reference point constraint then followed by correspondence correction using surface curvature feature. We tested the technique by applying thecorrespondence  correction technique into three registration algorithm registration which is ICP, ICP-AIF and ICP-SCF then compare it with the original registration algorithm The result shows that registration algorithm using correspondence correction 63% faster, 23% more accurate and to find 530% more correct pair matching point than the original registration algorithm.