Research On Global Point Cloud Registration Algorithm Based On Fractal Dimension

Point cloud registration,the basis of 3D modeling,is widely used in computer vision,medical modeling,track docking,workpiece inspection and cultural relic protection.The point cloud registration technology applied to production and life mostly requires accurate and fast.Therefore,the point cloud registration technology used in practice is usually non-automatic technology or multi-angle fixed multi-group scanning equipment.The automatic point cloud registration technology is still not mature in the corresponding point pair selection.Further research is needed between the partially overlapping point cloud data.A global point cloud registration algorithm based on fractal dimension was proposed,including coarse registration and fine registration.The main research contents are:(1)Point cloud data is discrete.In this paper,we apply the concept of two-dimensional fractal dimension to a three-dimensional discrete point cloud.First,calculate the fractal dimension value of each point in the point cloud,and extract the feature points according to the attribute of fractal dimension.(2)According to the distribution of feature points,the clustering algorithm is used to cluster the feature points to form multiple clusters.The relative structure between each cluster forms the global structure of the point cloud,which is represented by triangles.According to this structure,three pairs of corresponding point pairs are obtained.According to the corresponding point pairs obtained above,the point cloud coarse registration is completed.(3)There are some angle differences between the point cloud data obtained by the coarse registration algorithm.Therefore,the trimmed iterative closest point algorithm is used to complete the fine registration,and finally the point cloud registration process of this paper is completed.The experiments use classic point cloud data models.Experiments were carried out from two aspects: registration between point clouds with high overlapping rate and registration between point clouds with low overlapping rate.Compared with the current classical registration algorithms,the experimental results show that the proposed algorithm not only can register point cloud data with high overlapping rate,but also can register point cloud data with low overlapping rate.Using the standard model data of the computer graphics laboratory of Stanford University,the registration error of this paper is reduced by up to two orders of magnitude compared with the Go-ICP algorithm.The algorithm of this paper provides a novel solution for point cloud registration,which provides a new idea for point cloud registration.