【作者】 丁洁；

【导师】 裴海龙； 张宇；

【作者基本信息】 华南理工大学 ， 控制工程（专业学位）， 2016， 硕士

【摘要】 机载激光扫描系统(Light Detection and Ranging,LiDAR)作为一种主动式对地观测系统,具有自动化程度高、扫描速度快、受天气条件影响小等特点且能够获取到高密度、高精度的三维点云数据,因此在数字城市建设、地形测绘、海岸带监测等领域受到极大关注。而随着新应用领域的出现和新技术的快速发展,点云本身发生了巨大变化,主要体现在数据规模海量化、数据结构多样化等方面。现有的点云管理软件如Cyclone6.0、Polywork和Geomagic等,都侧重于解决点云建模问题,对于海量点云支持不足,主要原因是这些软件不具备高效的点云组织管理方法。针对以上问题,本文以海量非结构化机载LiDAR点云数据作为研究对象,以其存储方式为突破口,重点探索点云的组织管理。主要研究内容和创新点如下:(1)使用MongoDB管理机载LiDAR点云数据。以往基于文件和关系型数据库的点云管理系统,在海量数据的存储、访问、非结构化数据管理和扩展性等方面都存在难以克服的问题。本文应用MongoDB解决了海量机载LiDAR点云存储去结构化的问题,并充分挖掘其性能,以指导点云空间索引的设计工作。(2)设计了海量点云基于MongoDB的多层索引,其性能超越了多数基于内存的索引方法。首先以网格划分点云的二层空间并进行Hash编码,使用MongoDB的B树索引对二级索引码进行组织,形成基于空间网格的Hash码与B树的混合索引方法。考虑到应用中点云区域查询优先级高于单点查询的事实和Mongo DB的查询特性,利用JSON格式对点云建立了第三层存储结构,设计了点云基于MongoDB的多层索引方法,提高了区域查询效率。通过与传统索引方式的对比实验,证明此法具有更高的点云查找性能。(3)优化海量点云可视化性能。在完成了点云存储和索引的基础上,综合利用基于视点的数据裁剪、“5-5-5”动态调度策略、LOD技术、多线程和OpenGL等技术,实现了海量点云在普通PC上的高效检索和实时快速可视化,保证了海量点云漫游的流畅性和平滑性。在上述理论和实践研究的基础上,选择Linux Ubuntu14.04作为开发平台,利用C语言和Mongo-C-Driver1.2.1,结合OpenGL图形库,开发了海量机载LiDAR点云数据可视化系统,验证了本文点云数据组织管理方法的有效性和优越性。更多还原

【Abstract】 Airborne LiDAR is a kind of active earth observation system with a high degree of automation, scanning speed, and rarely influenced by weather conditions, etc. In addition, this system can obtain high precise, high dense three-dimensional terrain information directly. It is widely used in many fields such as digital city construction, topography mapping, coastal monitoring. With the emergence of new areas and the rapid development of new technologies, point-cloud data has changed a lot with the larger scale and uncertain data structure. The existing point-cloud data management softwares like Cyclone6.0, Polywork and Geomagic usually emphasis on the modeling and are lack of support for massive point-cloud data processing because of its inefficient organization and management of point-cloud data.To deal with these problems, this paper choose the massive unstructured airborne LiDAR point-cloud data as the research object, analyses its storage methods first and study on its organization and management. In this paper, the main research work and innovation are as follows:(1) Use MongoDB to manage point-cloud data. The past organization methods of point-cloud data based on the file or relational database storage are unable to deal with the storage and access of huge data, unstructured data management and expansibility, etc. This paper use MongoDB to store massive point-cloud data with uncertain data structure and excavate its property to guide the later study.(2) Design the multilayered spatial index of the massive point-cloud data based on MongoDB and it perform better than many other methods based on RAM. Firstly, split the point-cloud space twice with Hash code, then use B-tree index of MongoDB to arrange the second index code to generate the mix spatial index base on Hash code of 3DGrid and B-tree. Considering that area query efficiency always more important than point query efficiency in the application, use JSON to design the third index of point-cloud data to improve the area query efficiency and finally finish the multilayered spatial index based on MongoDB of the massive point-cloud data. Comparing to the traditional methods, this method is more efficiency.(3) Optimize the visualization performance of massive point-cloud data. On the basis of the massive point-cloud data organization and spatial index, this paper use data clipping, “5-5-5” dynamic schedule strategy, multithread and LOD technology to achieve visualization of massive point-cloud data using optimal scheduling approach with good continuity and smoothness in process of roaming.With the above obtained results in theory and practice, a visualization prototype system for massive airborne LiDAR point-cloud data is developed with MongoDB,OpenGL graphic library and C program language on the platform of Linux Ubuntu14.04. The effectiveness and superiority of the organization and management algorithms proposed in this paper are shown by the seamless running of the system developed.更多还原