基于三维激光扫描的点云数据逆向重建算法研究

发布时间：2019-07-05 19:26

【摘要】：近年来,随着三维激光扫描技术的发展,被广泛应用于测量学、工业控制等领域。由于点云数据有存储方便,计算灵活的特点,其已经成为计算机图形学中一种重要的元数据形式。而点云逆向重建是点云计算中的一项重要技术,随着激光扫描技术的发展以及扫描对象曲面的复杂化,其海量点云数据又为点云数据逆向重建带来了新的挑战。本文在详细研究了近几年点云数据逆向重建技术的基础上,针对海量的、杂乱的点云数据在逆向重建过程中存在的若干关键问题展开研究。首先,利用三维激光扫描仪实现海量点云数据获取,并以点云数据形式存储被测对象表面信息。其次,对点云数据进行不确定性分析、区域分割、数据精简等预处理操作。最终,依靠曲面重建算法实现被测对象的数据反演。具体研究工作内容如下:(1)研究点云数据的不确定性表征重点研究在实际测量过程中,扫描设备误差以及环境等因素与测量数据不精确的关系。为了能够量化原始测量数据的不精确度,并令其参与到点云数据的计算过程中,本文重点研究点云不确定性的表示模型,进而研究如何利用贝叶斯推理技术,量化整个测量空间点云数据的不确定度。(2)研究点云数据聚类分割算法在海量点云数据的分割过程中,聚类算法的执行效率成为系统实施的瓶颈。面对海量的点云数据,本文以K-means聚类算法为基础,研究点云数据的区域分割技术。为了有效地提高聚类算法的执行效率,减少算法迭代次数,本文引入了点云密度的概念,在K-means聚类簇的基础上,研究簇密度的估算模型以及簇中心的调整方法。(3)研究保留特征的点云数据精简算法为了压缩点云数据量,提高三维曲面建模的速度,本文研究保留曲面特征的点云数据精简算法。从点云数据的局部微分几何特征出发,以自然二次曲面作为点云数据模型,研究点云数据模型与点云数据之间的匹配关系;在点云数据与点云模型匹配关系的基础上,依据点云模型的曲面特征,研究如何分层次的实现保留几何特征的点云数据精简的方法。(4)研究点云曲面重建算法为了能够精准地完成点云曲面重建工作,本文研究泊松曲面重建算法。八叉树的深度值直接影响泊松曲面方程的执行效率以及曲面重建的效果,从而使得泊松曲面重建在孔洞细节处的重建效果不理想。本文在点云聚类分割的基础上,分析容易出现孔洞处的数据特征,研究如何结合贪婪三角化方法弥补泊松重建法在细节处重建效果差的现象。最后,本文利用悬臂式大车、三维激光扫描仪搭建了实验平台,研究适合料场数据反演的三维点云逆向重建系统的实施方法。针对获取的料堆数据集,验证了数据采集、数据预处理,进而实现堆料信息的反演。
文内图片：

图片说明：边缘数据区域选取示意图
[Abstract]:In recent years, with the development of three-dimensional laser scanning technology, it has been widely used in surveying, industrial control and other fields. Because point cloud data has the characteristics of convenient storage and flexible calculation, it has become an important form of metadata in computer graphics. Point cloud reverse reconstruction is an important technology in point cloud computing. With the development of laser scanning technology and the complexity of scanning object surface, its massive point cloud data brings new challenges to point cloud data reverse reconstruction. In this paper, based on the detailed study of the reverse reconstruction technology of point cloud data in recent years, some key problems existing in the process of reverse reconstruction of massive and chaotic point cloud data are studied. Firstly, 3D laser scanner is used to obtain massive point cloud data, and the surface information of the measured object is stored in the form of point cloud data. Secondly, the point cloud data uncertainty analysis, region segmentation, data simplification and other preprocessing operations. Finally, the data inversion of the measured object is realized by surface reconstruction algorithm. The specific research contents are as follows: (1) the uncertainty representation of point cloud data focuses on the relationship between scanning equipment error, environment and other factors and the inaccurate measurement data in the actual measurement process. In order to quantify the inaccuracy of the original measurement data and make it participate in the calculation of point cloud data, this paper focuses on the representation model of point cloud uncertainty, and then studies how to quantify the uncertainty of the whole measurement space point cloud data by using Bayesian reasoning technology. (2) in the process of massive point cloud data segmentation, the point cloud data clustering segmentation algorithm is studied. The execution efficiency of clustering algorithm has become the bottleneck of system implementation. In the face of massive point cloud data, this paper studies the region segmentation technology of point cloud data based on K-means clustering algorithm. In order to effectively improve the execution efficiency of clustering algorithm and reduce the number of iterations of the algorithm, the concept of point cloud density is introduced in this paper. on the basis of K-means clustering, the estimation model of cluster density and the adjustment method of cluster center are studied. (3) in order to compress the amount of point cloud data and improve the speed of 3D surface modeling, the point cloud data reduction algorithm with reserved features is studied in this paper. Based on the local differential geometric characteristics of point cloud data, the matching relationship between point cloud data model and point cloud data is studied by using natural Quadric surface as point cloud data model. On the basis of the matching relationship between the point cloud data and the point cloud model, according to the surface characteristics of the point cloud model, this paper studies how to simplify the point cloud data with geometric features at different levels. (4) in order to accurately complete the point cloud surface reconstruction, this paper studies the Poisson surface reconstruction algorithm. The depth value of octree directly affects the execution efficiency of Poisson surface equation and the effect of surface reconstruction, so that the reconstruction effect of Poisson surface reconstruction at hole details is not ideal. In this paper, on the basis of point cloud clustering segmentation, the data characteristics of holes are analyzed, and how to make up for the poor reconstruction effect of Poisson reconstruction method in detail is studied by combining greedy triangulation method. Finally, the experimental platform is built by using cantilever car and three-dimensional laser scanner, and the implementation method of three-dimensional point cloud reverse reconstruction system suitable for material field data inversion is studied in this paper. Aiming at the acquired data set, the data acquisition and data preprocessing are verified, and then the inversion of stacking information is realized.
【学位授予单位】：燕山大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN249

【相似文献】