Study On Data Registration And Reduction For 3D Point Clouds

As an advanced manufacturing technique, reverse engineering, which can be applied to product design, development and innovation, has become an independent subject in Computer Aided Design. Data processing, one of the key techniques in revere engineering, mainly covers data registration, data reduction and other techniques. Because the quality and the number of 3D data point have vital effect on the quality of subsequent model reconstruction, it's very significant to study on data registration and reduction. The concept and development of reverse engineering are presented in this thesis, and 3D point clouds registration and reduction are researched in detail. In this thesis, the contents of the research are as follows:1. A high-accuracy fine registration method is presented for partially overlapping point clouds that have been coarsely registered into a common coordinate system. The proposed algorithm, which is called dual interpolating point-to-surface ICP algorithm, is principally a modified variant of point-to-surface ICP algorithm. The original correspondences are established by adopting a dual surface fitting approach using B-spline interpolation. A novel"auxiliary pair"constraint, together with surface curvature information, is employed to remove unreliable matches. The combined constraint directly utilizes global rigid motion consistency in conjunction with local geometric invariant to reject false correspondences precisely and efficiently. For the two refined corresponding datasets, the least-square method is employed to solve the transformation parameters.2. According to the demand of actual applications as well as the properties of both 3D grid method and curvature method, a novel secondary reduction method is presented in this thesis. This algorithm is the combination of non-uniform grid method, which is based on the octree structure, and surface curvature method, which is based on B-spline surface fitting.3. Based on the fine registration and reduction algorithms proposed in this thesis, an integrated software system is developed using Microsoft Visual C++6.0 and OpenGL on the Windows XP operation system.As the experimental results demonstrate, for data registration, the ICP variant presented in this thesis can efficiently reject false correspondences to obtain high-quality refined correspondences, thus greatly improve the accuracy of fine registration for 3D point clouds. The proposed method not only can achieve small mean and standard deviation of registration errors, but also can fulfill high mutual interpenetration and uniform error distribution in the overlapping region after registration. For data reduction, the novel secondary algorithm not only can preserve critical surface characteristics and achieve high reduction rate, but can improve the operation efficiency compared with solely using curvature method.The innovation of this thesis: in data registration part, presenting an improved point-to-surface ICP algorithm that is based on B-spline surface fitting; presenting a novel"auxiliary pair"constraint and combining it with curvature constraint to remove unreliable correspondences. In data reduction part, an efficient secondary algorithm is presented that utilizes the merits of non-uniform grid method and surface curvature method.