Research On 3D Point Cloud Data Processing And Segmentation Algorithm

With the rapid development of computer vision,unmanned driving,3D city,intelligent robot and other fields in recent years,3D point cloud data has been widely used in these fields due to its unique spatial structure advantages.High-quality and fast reconstruction,and effective segmentation of the original point cloud data acquired by the scanning device is the key part of the current 3D point cloud data processing technology.Effective preprocessing of point cloud can greatly improve the speed and precision of reconstruction.High-quality segmentation is the basis of machine vision technology.This dissertation mainly studies the simplification,smoothing,reconstruction,registration and segmentation of point cloud data as follows:(1)The redundancy,noise and hole in the acquisition of point cloud data would affect the rapid and high-quality reconstruction.To solve this problem,a new point cloud reconstruction method is proposed.The method is based on the preliminary denoising point cloud data.Firstly,the point cloud data was simplified by VoxelGrid.And than using the improved moving least squares to make point cloud smooth.Finally using a multilateral local projection triangulation to rapidly reconstruct the point cloud.The experimental results show that the proposed method can effectively repair the local cavity of the point cloud and obtain a smooth manifold surface,which makes the reconstruction result closer to reality,and finally achieves high-quality reconstruction of point cloud data.(2)In the fine registration process of point cloud,Iterative Closest Points(ICP)is commonly used in point cloud registration due to the high registration precision.But the traditional ICP algorithm takes longer runtime with the number of iterations,and the nearest point pair with large deviation would affect the convergence of the error.On this basis,it is proposed that establishing quickly the topological structure of the point cloud data by the kd-tree,speeding up the search speed of the point cloud,and judging the nearest point pair with large deviation by the Euclidean distance judgment.The experimental results show that the proposed method is optimized in terms of iteration number,running time and registration error compared with the traditional ICP algorithm.(3)The segmentation of point cloud play a key role in the processing of point cloud data,regional growth is widely used in 3D point cloud segmentation,however,the uncertainty of the point cloud characteristics and the unreasonable selection of seed pointwill result in the instability of the traditional regional growth method.To resolve this problem,this paper presents an improved method of regional growth segmentation,we set the minimum curvature point to the seed point by estimating the magnitude of the curvature of point cloud data,then the growth criteria is determined according to the local characteristics of point cloud data.Experimental results show that this method can not only divide the point cloud data effectively,but also solve the problem of the instability of the traditional regional growth,improving the accuracy and reliability of point cloud segmentation.There are 32 figures,8 tables and 72 references in this paper.