基于GPGPU的并行LiDAR点云滤波算法  

A point cloud filtering algorithm based on GPGPU parallel computing

作　　者：崔放[1,2] 徐宏根[2] 王宗跃[3] 何文熹[2] 

机构地区：[1]中国地质大学(武汉)资源学院 [2]中国地质调查局武汉地质调查中心 [3]集美大学计算机工程学院

出　　处：《华中师范大学学报（自然科学版）》2014年第3期431-435,共5页Journal of Central China Normal University：Natural Sciences

基　　金：国家863计划项目(2012AA12A308);国家自然科学基金项目(41201462)

摘　　要：基于边缘点进行LiDAR点云滤波算法在复杂城区具有良好效果,然而在边缘点检测时非常耗时,因此提出基于GPGPU的并行边缘点检测算法来提高滤波的效率.该并行边缘点检测算法分为并行检索k最近邻点、并行拟合平面、并行计算投影点和并行计算夹角等4个步骤.在计算检索k最近邻算法中,提出了队首优先的插入排序算法,大量减少判断次数.实验结果表明,采用GPU并行计算有效提高滤波效率,当数据量达到128M以上,加速比可以达到9左右.The LiDAR point cloud filtering algorithm based on marginal points,is performed well in complex cityscapes.However,it is very time consuming for detecting marginal points.To improve the efficiency for LiDAR point cloud filtering,a parallel marginal point detecting algorithm is proposed based on GPGPU.The algorithm is followed by four steps：parallel indexing K nearest neighbor points,parallel plane fitting,parallel calculating projection points and parallel calculating points＇ angle.For the step of parallel indexing k nearest neighbor points,an insertion sorting algorithm based on queue priority is proposed to reduce the number of judging amount substantially.The experiment results show that GPU parallel computing can improve the LiDAR point cloud filtering efficiency effectively.The acceleration ratio can reach about 9 when data size is above 128 megabytes.

关 键 词：LIDAR 滤波 GPGPU 

分 类 号：P237[天文地球—摄影测量与遥感]