面向工程计算的岩体三维重建理论与方法研究

3D reconstruction of rock mass for engineering computation forms the basis of geological visualization and numerical analysis.With the popularity of modern non-contact measurement equipment for geological survey, 3D reconstruction of rock mass based on point cloud has been a research hotspot in the field of rock mass 3D reconstruction, which is faced up with several problems. Research is conducted on the following 4 aspects in this project for these problems: (1) a reliable and robust vegetation filter algorithm for rock mass point cloud is designed with the analysis of occlusion problem for the rock mass surface caused by vegetation by utilizing the feature extraction methods based on semantics and the machine learning theory, and a corresponding hole repair algorithm is proposed in order to solve the hole problem caused by filtering by referencing the image repair theory; (2) an effective registration model for rock mass point cloud is established based on a minimum registration set aiming at the fast registration problems of massive point cloud; (3) a parallel model of triangulation for massive point cloud based on hybrid indexing method is studied and constructed; (4) an automatic discontinuities extraction algorithm based on point cloud process and image process and a 3D reconstruction algorithm of rock mass based on coupling mechanism are studied and designed, based on which a rock mass 3D reconstruction primitive system for engineering computation is implemented. Achievements of this research have important meanings in sciences and applications, and can be directly used in the computation of engineering and sciences.

面向工程计算的岩体三维重建是工程地质可视化和数值分析的重要基础。随着现代非接触测量设备在地质勘测领域的普及，基于点云的岩体三维重建已逐渐成为岩体三维重建领域的研究热点，并面临诸多亟待解决的问题，本项目将针对这些问题在以下四个方面展开关键技术研究：（1）利用基于语义的特征提取方法和相关机器学习理论等分析植被对岩体表面的遮挡问题，设计可靠鲁棒的岩体点云植被滤波算法，并借鉴图像修复理论分析滤波后导致的空洞问题，设计对应的空洞修复算法；（2）针对大规模海量点云的快速配准问题，建立基于最小配准集合的岩体点云高效配准模型；（3）研究并建立基于混合索引方法的海量点云三角化并行模型；（4）研究设计结合点云和图像处理的结构面自动提取算法和基于耦合机制的岩体三维重建算法，并据此完成面向工程计算的岩体三维重建原型系统。研究成果具有重要的科学意义和应用价值，并可直接应用到工程和科学计算中。

面向工程计算的岩体三维重建是工程地质可视化和数值分析的重要基础。本项目按照计划对岩体三维重建的关键技术展开了相关研究，实现了预期目标。项目分析了岩体点云的特点，构建了有效表征岩体特征的维度特征，并基于此特征，提出了2个基于神经网络模型的岩体点云滤波算法，有效提升了岩体点云的滤波效果；针对岩体中不规则平面较多的特点，提出了4个性能良好的岩体点云平面提取算法，分别是基于霍夫变换和区域生长的多尺度平面提取方法、基于随机采样一致性和区域生长的平面提取方法、基于主方向估算的岩体点云平面提取方法和基于多尺度区域生长的岩体点云平面提取方法，有效解决了岩体平面提取中的过分割和欠分割问题；基于提出的平面提取算法和岩体点云特点，设计了基于平面提取的配岩体点云配准算法，同时针对特征提取困难等问题，分别提出了基于N阶完全图的高效岩体点云配准算法和面向无靶标辅助的岩体点云快速配准算法，有效提升了岩体点云配准精度和效率；基于岩体点云的邻域特征，提出了自适应的孔洞修复算法，该算法在修复孔洞的同时，较好地保留了岩体自身的拓扑结构；针对岩体点云的海量数据特点，设计了采用分层结构的混合索引方式，并针对海量点云三角化并行实施中的边界问题，设计了两个边界处理方案；基于所提出的各种算法，设计并开发了一个面向工程计算的岩体点云三维重建系统，实现了岩体点云三维重建的整个流程，可直接对岩体点云进行各种处理并最终重建三维模型。项目研究成果促进了面向工程计算的三维重建技术的深入研究，并可服务于工程和科学计算。