大规模工程点云数据的归并式八叉树管理及可视化  

Merging-based hierarchical octree management and visualization for large-scale engineering point clouds

作　　者：张宏阳 金银龙[1] 张礼兵 刘全[1] 王浩[1] 庞小荣 ZHANG Hongyang;JIN Yinlong;ZHANG Libing;LIU Quan;WANG Hao;PANG Xiaorong(State Key Laboratory of Water Resources Engineering and Management,Wuhan University,Wuhan 430072,Hubei,China;PowerChina Kunming Engineering Corporation Limited,Kunming 650051,Yunnan,China)

机构地区：[1]武汉大学水资源工程与调度全国重点实验室,湖北武汉430072 [2]中国电建集团昆明勘测设计研究院有限公司,云南昆明650051

出　　处：《水利水电技术(中英文)》2024年第12期159-170,共12页Water Resources and Hydropower Engineering

基　　金：云南省科技厅重大科技专项计划(202202AF080003);国家自然科学基金项目(51879207);西藏自治区清洁能源科技重大专项(XZ202201ZD0003G01)。

摘　　要：【目的】随着智慧水利体系建设的不断推进,三维激光扫描技术因其全面监测、扫描速度快、精度高的技术特性备受关注。然而,高冗余、非结构化、空间分布不均匀等点云特点给超大规模工程点云数据管理和可视化带来了极大的挑战。针对传统的“自上而下,逐级细分”的点云索引构建方法不能很好适配海量点云分块处理或多点云输入场景的问题,提出一种“自下而上,由小及大”的点云八叉树索引构建策略及海量点云快速渲染方法。【方法】结合点云全局信息建立全局格网索引和索引编码转换体系,通过多线程并行读取点云数据进行逐点分配、格网归并等操作,实现海量点云数据的八叉树索引快速构建。此外,采用非冗余采样策略建立多层次细节模型和多线程动态调度技术,实现海量点云数据的高质量可视化渲染。【结果】试验结果显示,基于归并构建策略的方法可处理超50亿点规模的点云数据,索引构建过程内存消耗峰值为2.78 GB,处理效率可到200万点/s。与点云库(Point Cloud Library,PCL)、CloudCompare等点云处理工具相比,在内存资源利用、索引构建效率、点云渲染方面表现出良好的性能。【结论】大规模工程点云数据管理和可视化是深化工程点云监测全流程研究的基础问题,其关键在于优化超大体积点云数据的空间索引快速构建。“先细分,后归并”的处理策略可有效缓解分块数据空间重叠冲突问题,算法逻辑清晰,尤其适用于四叉树、八叉树等基于点云包围盒空间剖分映射的索引结构。[Objective]With the continuous advancement of intelligent water conservancy,3 D laser scanning technology has attracted significant attention due to its comprehensive monitoring capabilities,rapid scanning speed,and high precision.However,these point clouds,characterized by massive scale,unstructured nature,and uneven density,present formidable challenges in engineering data management and visualization.Aiming at the limitations of the“top-down,step-by-step subdividing”point cloud indexing construction method,which are not well-suited for massive point cloud chunking or multi-point cloud input scenarios,a“bottom-up,small-to-large”octree construction method for massive point data management and fast point cloud rendering was proposed.[Methods]First,a global voxel-based indexing and encoding conversion system is established based on the global information of point clouds.This framework facilitates the rapid octree indexing for massive point cloud data through operations such as point-by-point allocation and grid merging,which are executed in parallel by multi-threaded processing.Furthermore,a non-redundant sampling strategy is adopted to build the LOD(level of details)models,coupled with multi-threaded dynamic scheduling technology,to achieve high-quality visualization and rendering of massive point clouds.[Results]Experimental result demonstrated that the merge-based construction strategy could handle point cloud data exceeding 5 billion points.The peak memory consumption during the indexing process was 2.78 GB,and the processing efficiency reached 2 million points per second.Compared to point cloud processing tools like Point Cloud Library(PCL)and CloudCompare,the proposed method exhibited superior performance in memory management and indexing struct construction efficiency,as well as point cloud rendering.[Conclusion]The management and visualization of large-scale engineering point cloud data are fundamental to advancing the fullprocess research of engineering point cloud monitoring.The key lies in optim

关 键 词：三维激光扫描 海量点云 八叉树构造 内外存动态调度 归并算法 多层次细节模型 水利工程 数字孪生 

分 类 号：TV512[水利工程—水利水电工程] TP391[自动化与计算机技术—计算机应用技术]