Research On Point Cloud Registration Technology Of Intraoral Scanning System

The intraoral scanning system applies 3D reconstruction technology to the field of oral medicine,turning the traditional complicated and time-consuming manual impression process into a fast and accurate digital one.As an important step in 3D reconstruction,point cloud registration is to transform the point clouds generated by photos from different perspectives into the same coordinate system,so that the corresponding positions in different point clouds overlap each other,thereby restoring the geometric information of the object completely.The intraoral scanning system requires high reconstruction accuracy,and inaccurate point cloud registration will greatly reduce the final reconstruction accuracy.Therefore,how to design an accurate point cloud registration process is the research focus of this thesis.The main research content of this thesis includes:(1)Point cloud denoising method based on point cloud segmentation and coordinate constraints.There are teeth and gums in the point clouds of intraoral scanning,and the existence of the gums will affect the registration accuracy of the teeth.Therefore,this thesis incorporates point cloud segmentation into the point cloud denoising process.First,the region growing method is used to divide the point cloud into disjoint regions.And then,based on the characteristic that the teeth part in the point cloud is closer to the viewpoint,the vertices closest to the viewpoint in each region of the point cloud are calculated,and finally the valid coordinate range and the viewpoint closest constraint are utilized to filter out the tooth part in the point cloud.Experiments show that the proposed denoising method can increase the registration recall rate by20.3% on average.(2)The curvature-based interval sampling method.Generate a statistical histogram according to the curvature of the vertices,and then perform random sampling in the histogram interval,which can solve the time-consuming problem of the normal space sampling.Experiments show that the algorithm can provide registration accuracy similar to that of voxel grid and normal space sampling,and the speed is increased by an average of 31.9% compared to normal space sampling.The speed advantage becomes more pronounced as the number of samples increases.(3)According to the characteristics of point clouds of intraoral scanning,a weighted error measurement based on viewpoint distance and a fusion distance constraint for the Iterative Closest Point(ICP)algorithm are proposed.Traditional ICP registration uses Euclidean distance as a single error measure,which may cause inaccurate registration in some cases.For this reason,this thesis combines the Fast Point Feature Histogram(FPFH)feature distance with the Euclidean distance.First,the Euclidean distance and the FPFH histogram distance of the corresponding point are calculated,and then the normalized two distances are used to construct the fusion distance,and finally only the parts with the shortest fusion distance are reserved to participate in the error calculation.Experiments show that the combined use of the two strategies can increase the registration recall rate of ICP registration by 7.1% on average.(3)Design and implement an intraoral scanning system software.Based on the proposed denoising method,down-sampling method and improved ICP method,combined with the overall process of intraoral scanning,an intraoral scanning system software is designed.And the implementation of preprocessing module,registration module and meshing module is introduced.Tests show that the three modules work properly and output the expected results.