基于外存八叉树的三维激光点云实时渲染技术研究

发布时间：2018-05-06 16:26

本文选题：实时渲染 + 八叉树　；参考：《天津师范大学》2017年硕士论文

【摘要】：三维点云数据的交互式处理通常依赖点云数据三维实时渲染。随着三维激光扫描技术的深入应用,越来越多的三维点云数据被获取和积累下来。随着点云数据量的不断增大,大规模三维激光扫描点云的实时渲染已经成为点云数据处理的瓶颈问题。本文提出一种支持大规模点云实时渲染的技术方法,主要成果如下:根据用RTK测得的扫描站点和靶标的真实地理坐标与其对应的在激光扫描仪独立坐标系下的相对坐标,使用坐标转换矩阵经旋转和平移将点云数据全部统一到真实地理坐标系中,然后通过配准完成多站地面三维激光扫描点云的融合。研究了点云数据的数据特征,采用"分之而治"的思想,使用C++语言实现了对大规模点云数据的分块,每个分块包含一定数量的扫描点,可直接读入内存进行处理。对于每个点云分块,构建kd tree,使用PCL滤波以最近邻点数为阈值滤除点云中的孤立点。遍历经过去噪的点云分块,统计测区覆盖的空间范围。在外存储器上建立不同细节层次的八叉树索引结构,对海量点云数据进行有效的组织。通过与商业软件的对比渲染实验,验证了本文方法的可行性。最终的实验结果表明,本文提出的方法支持对超过内存容量的三维点云数据进行实时渲染。该方法有望用于大规模三维激光扫描点云数据的可视化和交互式处理。
[Abstract]:The interactive processing of 3 D point cloud data usually depends on 3 D real time rendering of point cloud data. With the further application of 3D laser scanning technology, more and more 3D point cloud data have been acquired and accumulated. With the increasing of point cloud data, the real-time rendering of large scale 3D laser scanning point cloud has become the bottleneck of point cloud data processing. This paper presents a technical method to support large-scale point cloud real-time rendering. The main results are as follows: according to the real geographical coordinates of scanning stations and targets measured by RTK and their corresponding relative coordinates in the independent coordinate system of laser scanner, The point cloud data are all unified into the real geographical coordinate system by rotation and translation of coordinate transformation matrix, and then the fusion of multi-station 3D laser scanning point cloud is completed by registration. In this paper, the data characteristics of point cloud data are studied, and the idea of "point and rule" is adopted, and the block of large scale point cloud data is realized by C language. Each block contains a certain number of scanning points, which can be read directly into memory for processing. For each point cloud block, kd tree is constructed, and PCL filter is used to filter the outliers in the point cloud with the nearest neighbor points as the threshold. Traversing the de-noised point cloud into blocks, statistics the coverage of the spatial range of the area. The octree index structure with different detail levels is built on the external memory to organize the massive point cloud data effectively. The feasibility of this method is verified by comparison with commercial software. The experimental results show that the proposed method can render 3D point cloud data over memory capacity in real time. This method is expected to be applied to the visualization and interactive processing of large-scale 3D laser scanning point cloud data.
【学位授予单位】：天津师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P225.2

【参考文献】