基于多尺度虚拟格网的LiDAR点云数据滤波改进方法  

Improved Method for LiDAR Point Cloud Data Filtering Based on Hierarchical Pseudo-Grid

作　　者：黄作维[1] 刘峰[2] 胡光伟[1] 

机构地区：[1]湖南工业大学建筑与城乡规划学院,湖南株洲412000 [2]中南大学地球科学与物理信息学院,湖南长沙410083

出　　处：《光学学报》2017年第8期346-355,共10页Acta Optica Sinica

基　　金：国家自然科学基金(43462378);湖南省自然科学基金(2017JJ2072)

摘　　要：机载激光雷达(LiDAR)点云数据滤波是目前点云数据处理领域研究的重点。针对目前点云数据滤波的难点,提出了一种基于多尺度虚拟格网和并行计算的改进滤波方法。该方法通过点云数据构建多级虚拟格网,对格网进行多尺度分解,剔除LiDAR数据中的粗差点,获取初始地面点及地物点;根据双向阈值滤波原理,以网格尺度由大到小的顺序逐层进行滤波处理,得到较为精细的地面点,并结合点云数据的并行编程处理,减少了滤波算法的误差积累。实验结果表明:改进算法与其他经典滤波算法相比,滤波精度有了较大的提高,在不同的地形条件下能有效地控制第II类误差,同时减少了总误差,提高了滤波处理的效率和数字高程模型(DEM)的可靠性。Point cloud data filtering of airborne light detection and ranging（LiDAR）is the focus in the current study of point cloud data processing field.In order to deal with the difficulty of point cloud data filtering at present,an improved filtering method based on hierarchical pseudo-grid and parallel computing is presented.In this method,hierarchical pseudo-grid is established by point cloud data,and the grid is multi-scale decomposed.The original gross error points of LiDAR data are eliminated.The ground point and planimetric point are obtained.According to the principle of double threshold filtering,more refined ground points are obtained by filtering process gradually with the order from big to small mesh scale.And the parallel programming process for point cloud data is combined to reduce the error accumulation of filtering algorithm.Experimental results show that the filtering accuracy of the improved algorithm is enhanced greatly compared to other classical filtering algorithms.The type II errors are controlled effectively in different terrain conditions.Meanwhile,the total errors are decreased,the efficiency of filtering process and the reliability of digital elevation model（DEM）are enhanced.

关 键 词：遥感 激光雷达 数据滤波 多尺度虚拟格网 并行处理 自适应阈值 

分 类 号：P237[天文地球—摄影测量与遥感]