海量点云通用图形处理器缓存机制与并行编辑方法  

Massive point cloud GPGPU buffer organization and parallel editing method

作　　者：金一杲 胡翰[1] JIN Yigao;HU Han(Faculty of Geosciences and Environmental Engineering,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]西南交通大学地球科学与环境科学学院,成都611756

出　　处：《测绘科学》2023年第7期200-207,共8页Science of Surveying and Mapping

摘　　要：针对现有海量点云数据组织常采用树索引结构,不支持被通用图形处理器(GPGPU)并行计算,无法实现处理结果与可视化的实时共享,难以满足实时点云编辑应用等问题,该文提出了海量点云GPGPU缓存组织与并行编辑方法,设计了一种基于GPU顶点缓存的海量点云数据组织方法,基于计算着色器技术实现了可视化数据与点云属性的直接共享与并行处理,满足实时点云选择、删除、查询、属性修改等编辑应用,同时采用操作栈的数据结构支持编辑操作的高效回退。实验结果表明,对于亿级大规模点云,该文方法相比传统基于空间索引结构的方法,在点云编辑效率上具有较明显的优势。Aiming at the problems that the existing massive point cloud data organization often uses tree index structure,which does not support GPGPU parallel computing,cannot realize real-time sharing of processing results and visualization,and is difficult to meet real-time point cloud editing applications,this paper proposes a massive point cloud GPGPU buffer organization and parallel editing method,designs a massive point cloud data organization method based on GPU vertex buffer,realizes direct sharing and parallel processing of visualization data and point cloud attributes based on computational shader technology,satisfies real-time point cloud selection,deletion,query,attribute modification and other editing applications,and adopts the data structure of operation stack to support efficient fallback of editing operations.The experimental results show that for large-scale point clouds of hundreds of millions,the method in this paper has obvious advantages in point cloud editing efficiency compared with the traditional method based on spatial index structure.

关 键 词：点云编辑 GPGPU OPENGL 计算着色器 

分 类 号：P237[天文地球—摄影测量与遥感]