Research On High-Precision Processing Technology Of Point Cloud Data For 3D Measurement With Swept Interference

3D point cloud processing is an important part of swept frequency interferometry measurement of 3D topographic.Its main task is to realize high-precision 3D point cloud automatic registration and surface reconstruction for point cloud scanned from multiple perspectives on the surface of the measured object without prior information.However,the existing point cloud processing methods cannot meet the needs of high-precision 3D topography measurement.Based on preprocessing,this paper studies the pairwise fine registration,multi-view registration and triangular reconstruction to improve the overall 3D reconstruction accuracy.The main research work of this paper is as follows:Firstly,describe the principle of swept frequency interferometry and the scanning principle of 2D turntable,and introduce the point cloud coordinate acquisition method of swept frequency interferometry 3D scanner.In addition,introduce the basic principles of 3D reconstruction technology,including KD-tree spatial index construction,through filtering and outlier filtering preprocessing,pairwise registration,multi-view registration and point cloud surface reconstruction.Secondly,in view of the fact that the misalignment between points in the overlapping area caused by the sparseness of the point cloud seriously affects the registration accuracy,this paper proposes a high-precision registration method based on the dual-constrained mutual projection between surfaces.The initial registration corresponding point set is constructed through the mutual projection between the feature descriptors in similar regions,and then the final registration point set is determined by the rigid transformation consistency constraint,which realize high-precision pairwise point cloud registration.Further,integrate K-means clustering,and propose a multi-view registration algorithm from the point to the local surface projection of the cluster center to achieve high-precision multi-view point cloud registration.In the verification experiment of the Stanford data set,under the same initial pose conditions,compared with the existing optimal algorithm,the pair-wise registration rotation error is reduced by 27.0%while keeping the translation error basically unchanged;the average rotation error of the multi-view registration result is reduced by 14.3%,and the average translation error is reduced by 4.8%.In the swept frequency interferometry experiment,a registration evaluation method that separates the measurement instrument error from the algorithm error is proposed.Under the same initial pose conditions,the pairwise registration result is comparable to the existing optimal algorithm;the average rotation error of multi-view registration is reduced by 32.8%,and the average translation error is reduced by 8.6%.Finally,the existing surface reconstruction algorithm is improved by randomly selecting multiple growth regions and moving least squares smoothing.A multi-point random region growth triangle mesh surface reconstruction algorithm is proposed to realize the anti-noise and anti-occlusion 3D surface reconstruction.On this basis,the surface deviation of the proposed registration method is evaluated,and the comparison between the proposed method and the existing method is given,the surface deviation of pairwise registration is reduced by 4.9%.The surface deviation of multi-view registration is reduced by 15.0%,and achieve high-precision swept frequency interferometric 3D measurement point cloud reconstruction.