Point Cloud Rigid Registration Method Based on Local Poisson Surface Reconstruction

Abstract

This paper provides a point cloud rigid registration method based on local sample reconstruction Poisson surface to find corresponding points, which belongs to the field of digital design and manufacturing, and is characterized in that: initial registration interactively selects feature point pairs in floating point cloud and fixed point cloud , construct a Poisson surface based on the neighborhood point set of fixed point cloud feature points; build a KD tree of the surface, query the nearest point of the floating point cloud sample point in the KD tree as a reference point, and use the sample point to the reference point ring domain patch The nearest point of the corresponding point is used as the corresponding point, and the measurement function is established based on the corresponding point pair, and the transformation parameters are solved by the SVD method; on the basis of the initial registration, the fine registration is based on the public domain to obtain the feature point pair adaptively, and the point to the Poisson surface is used. The nearest distance establishes an error measure to calculate the transformation parameters and further improve the registration accuracy. The invention can achieve higher registration accuracy for the initial registration, and the fine registration can quickly converge to the global optimum and has higher robustness.

Description

technical field [0001] The invention provides a point cloud rigid registration method based on partial Poisson surface reconstruction, which can be used for registration of multi-view point cloud data of sampling data on the surface of a physical object, and belongs to the field of digital design and manufacturing. Background technique [0002] In the fields of reverse engineering, computer graphics, quality inspection, etc., it is necessary to obtain 3D point cloud data from the surface of the object. Due to the occlusion of the surface of the object and the limitation of the measurement range of the scanning equipment, the current mainstream laser measurement equipment and grating projection measurement equipment must scan the measured object from multiple angles and sub-regions to obtain all point cloud data on the surface of the object. However, during the transformation angle scanning process, the coordinate system of the point cloud data obtained by each scan is differ...

AI Technical Summary

Problems solved by technology

Rusu et al. "Semantic 3d object maps for everyday manipulation in human living environments" (KI-Künstliche Intelligenz, 2010, 24(4): 345-348.) proposed feature descriptors based on FPFH for feature point matching, and added feature point discrimination conditions , the identified candidate feature points are further excluded by geometric information such as curvature and normal, and the unsuitable feature points are eliminated. The whole process uses the KD tree to speed up the search. This method has achieved a very good registration effect. Therefore, the The method is adopted by the open source point cloud algorithm library PCL, but this method still does not solve the drawbacks of excessive dependence on the initial relative position. As long as the initial relative position of the two point cloud models is not ideal, the registration will fail

Various improved methods improve the registration accuracy by finding more accurate geometric primitives and establish more efficient error estimation rules to increase the speed of convergence, but the problems brought about are complex calculations, high time costs, and the point clouds processed by each method Data is often specific and not universal

Images