Curvature-based Registration and Segmentation for Multisensor Coordinate Metrology

Abstract

With the rapid development of multiple sensors for shape acquisition and inspection, point-based discrete shape modeling is being widely used in many engineering applications, e.g. reverse engineering, quality control, etc. Geometry processing, which aims at recovering information about topology, geometry and shape from the measured data is one of the critical issues to achieve multiple sensors integration in coordinate metrology. This paper presents a novel approach for discrete geometry processing in multisensor coordinate metrology. Two important issues are addressed here: registration and segmentation. We propose here a new modified Iterative Closest Point (ICP) algorithm to improve the registration performances by using the curvature information. Shape recognition and segmentation are the most critical issues of discrete geometry processing. The local surface types and the characteristic points are first recognized based on two surface descriptors: shape index and curvedness. A clustering method is developed to classify the vertices according to their surface types, and a connected region generation approach is developed for final segmentation. Finally, an industrial case study is presented to illustrate the entire approach, and to demonstrate the validity of the proposed methods for engineering applications.