Moving Least Squares Surface Based Multi-view Point-data Registration

Multi-view point-data registration is not only the fundamental data process in reverse engineering, but also the key technology of multi-sensor based closed-loop measurement system. It is difficult to achieve better result with the existing methods for multi-view point-data registration as they are affected by the measurement noise and the complicated shape of the object measured. Therefore, it is valued for the research on multi-view point-data registration to some extent. The method of registration for multi-view point-data is studied systematically in this thesis, and the main researches include:1. The reconstruction of surface from the input point-data can improve the accuracy of the registration result, however, the segmentation and model selection are indispensible for the surface reconstruction, and both of them are still under study. Aiming at this problem, a method combining with Moving Least-Squares (MLS) surface is proposed. As a smooth manifold approximation from the point-data, the MLS surface needs no segmentation or model selection.2. The computation for orthogonal projection from point to MLS surface is the key technique for building the corresponding point relation which is indispensible to register the point-data in different views onto the MLS surface through the ICP registration. A numerical algorithm of second-order convergence is proposed in this thesis to implement the computation of the orthogonal projection efficiently. The osculating circle is built to locally approximated the intersected curve of the MLS surface in the algorithm to accelerate the convergence so as to improving the efficiency.3. Based on the research stated above, a method of MLS surface based multi-view point-data registration is proposed and implemented. Moreover, comparison and analysis are expanded between some classic method for multi-view registration and the method proposed in this thesis. The comparison shows that a better result of registration is obtained through the method proposed in this thesis.