三维点云数据的去噪和特征提取算法研究

发布时间：2018-05-14 01:35

本文选题：逆向重建 + 点云去噪　；参考：《南昌大学》2017年硕士论文

【摘要】：随着三维扫描技术、计算机技术以及信息技术的快速发展,通过实际物体扫描测量得到虚拟三维点云信息的模型重建技术已经成为了当前研究的热点。其应用领域十分广泛,例如虚拟现实中的模型构建、文物古迹的保护修复、3D游戏角色的开发以及各种模具的设计和修复等等。而在模型重建的过程中,点云模型预处理(包括去噪平滑、点云精简、点云特征提取以及点云拼接等)是三维逆向建模中非常重要的一部分。本文主要针对三维点云预处理环节中的点云去噪和点云特征提取算法进行研究,论文主要研究内容如下:(1)散乱点云邻域及微分信息估算:针对散乱点云的领域以及微分几何信息的估算进行理论研究。基于本文的散乱点云数据量较大的特点,固选用搜索效率较高的k-d树的搜索法对点云数据进行拓扑结构的建立。然后针对散乱点云的微分几何信息,首先根据曲面论相关内容对点云的法矢向量和点云的曲率进行研究,紧接着提出采用基于采样点邻域局部二次曲面的拟合方法对模型点云进行微分几何信息的估算。(2)基于特征信息分类的三维点数据去噪:针对三维散乱点云去噪逆向建模过程中,模型特征信息难以维持以及双边滤波对于特征信息少的区域容易产生过光顺的缺点,提出一种基于点云曲率特征信息分类的去噪算法。具体方法为采用局部曲面拟合的方法对点云的微分几何信息进行计算,由计算得到的模型点云的平均曲率设置局部特征权值。然后根据平均曲率特征权值对点云数据进行特征区域划分,得到特征信息较少的平坦区域和特征信息较多的区域。最后采用针对性的去噪算法,即邻域距离平均滤波算法和自适应双边滤波算法对划分好的两种不同特征区域分别进行去噪滤波。通过进行对不同去噪算法的对比实验和在不同噪声强度下的对比实验,结果表明本文算法具有较好的去噪效果。(3)散乱点云数据提取算法:针对散乱点云特征提取过程中,提取效率低和对噪声的敏感性差等问题。本文采用一种双阈值检测的特征提取算法。首先采用主成分分析法和局部曲面拟合的方法对散乱点云的微分几何信息(包括法向矢量和曲率信息)进行估算。然后根据计算出的邻域内法向矢量交角和平均曲率设定双重特征判定阈值,由此来对模型的点云数据进行特征提取。通过进行不同算法和不同噪声强度下的特征提取实验,结果表明本文算法具有较好的特征提取效果。(4)扫描和数据处理系统搭建:以手持式扫描仪和投影光栅式扫描仪为数据获取端,然后以C++为编程语言,利用OpenGL图形库和VS2013开发工具来实现对三维点云数据的去噪和特征提取的算法,并以汽车石膏模型为实例进行整个数据处理过程的说明,并进行算法可行性论证。
[Abstract]:With the rapid development of 3D scanning technology, computer technology and information technology, the model reconstruction technology to obtain virtual 3D point cloud information through actual object scanning measurement has become a hot research topic at present. It is widely used in many fields, such as model construction in virtual reality, protection and restoration of cultural relics and monuments, development of 3D game characters, design and restoration of various molds, etc. In the process of model reconstruction, point cloud model preprocessing (including denoising and smoothing, point cloud reduction, point cloud feature extraction and point cloud stitching) is a very important part of 3D reverse modeling. In this paper, the point cloud denoising and point cloud feature extraction algorithms in the preprocessing of 3D point cloud are studied. The main contents of this paper are as follows: (1) neighborhood and differential information estimation of scattered point clouds: the field of scattered point clouds and the estimation of differential geometric information are studied theoretically. Based on the large amount of scattered point cloud data in this paper, the search method of k-d tree with high search efficiency is used to establish the topological structure of point cloud data. Then according to the differential geometry information of scattered point cloud, the normal vector and curvature of point cloud are studied according to the related content of surface theory. Then, a fitting method based on the local Quadric surface of sampling points is proposed to estimate the differential geometric information of the model point cloud. (2) based on the classification of feature information, 3D point data denoising is proposed: in the process of inverse modeling for 3D scattered point cloud denoising, Model feature information is difficult to maintain and bilateral filtering is easy to produce over-fairing for regions with less feature information. A denoising algorithm based on point cloud curvature feature information classification is proposed. The specific method is to calculate the differential geometric information of the point cloud by using the method of local surface fitting, and the local eigenvalue is set from the average curvature of the model point cloud. Then the point cloud data are divided according to the mean curvature feature weights, and the flat region with less feature information and the region with more feature information are obtained. Finally, two kinds of de-noising algorithms, namely neighborhood distance average filtering algorithm and adaptive bilateral filtering algorithm, are used to Denoise the two different characteristic regions. Through the contrast experiment of different denoising algorithm and the contrast experiment under different noise intensity, the result shows that the algorithm has better denoising effect. 3) the algorithm of extracting scattered point cloud data: in the process of feature extraction of scattered point cloud, The extraction efficiency is low and the sensitivity to noise is poor. In this paper, a feature extraction algorithm based on double threshold detection is proposed. Firstly, the differential geometric information (including normal vector and curvature information) of scattered point cloud is estimated by principal component analysis and local surface fitting. Then the threshold value is set according to the normal vector intersection and mean curvature in the neighborhood, and then the point cloud data of the model is extracted. Through the experiments of feature extraction under different algorithms and different noise intensity, The results show that the algorithm has a better feature extraction effect. 4) scan and data processing system is built: the hand-held scanner and projection raster scanner are used as the data acquisition terminal, and then C is used as the programming language. The algorithm of de-noising and feature extraction of 3D point cloud data is realized by using OpenGL graphics library and VS2013 development tool. The whole data processing process is illustrated by taking the automobile gypsum model as an example, and the feasibility of the algorithm is demonstrated.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.7

【相似文献】