In this paper, a novel approach to achieve model reconstruction based on global features is proposed on the ground of our research on multi-view scans registration. The algorithm combines the rigid registration and non-rigid registration. Principle component analysis (PCA) is adopted to achieve the rough pair-wise registration. Then, we do some research on the features selection and feature correspondence and propose an improved method to achieve fine pair-wise registration based on Iterative Closest Points (ICP). Finally, an effective global features non-rigid registration is proposed based on thin-plate splines (TPS), which overcomes both the accumulation errors generated in the process of pairwise registration and errors induced by laser devices. The feasibility and usability of the algorithm are confirmed through related experiments.
展开▼