倾斜地面3D点云快速分割算法  

A Fast Segmentation Algorithm for 3D Point Cloud on Inclined Ground

作　　者：张清宇 崔丽珍[1] 李敏超 马宝良 ZHANG Qingyu;CUI Lizhen;LI Minchao;MA Baoliang(School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010

出　　处：《无线电工程》2024年第2期447-456,共10页Radio Engineering

基　　金：国家自然科学基金(62261042);内蒙古自治区科技计划项目(2022YFSH0051)。

摘　　要：依赖于平面拟合或局部几何特征区分地面障碍物的方法广泛应用于自动驾驶领域,但是在具有倾斜地形或稀疏数据的情况下,性能会降低。在倾斜地形场景中,利用激光雷达(LiDAR)提供的点云数据,通过地面点云的3D投影与直线拟合,利用栅格地图方法降低计算复杂度,实现地面点云的分割。针对非地面的点云集合,利用SLR聚类算法处理,通过设定强度特征阈值在垂直方向区分地面点与非地面点,并对扫描到的障碍物地面分类。通过实验分析,提出的算法较其他地面点云分割算法,一方面在倾斜地形上具有更好的建图效果,另一方面SLR聚类算法处理后的强度特征在X、Y、Z三个方向覆盖范围更精确,如在X方向相较于快速地面分割算法平均提高了44.0%,相较于添加了栅格地图的算法平均提高了40.1%。The methods that rely on plane fitting or local geometric features to distinguish ground obstacles are widely used in the field of automatic driving,but their performance could be reduced in the case of sloping terrain or sparse data.In a sloping terrain scene,the point cloud data provided by LiDAR,the grid map method through 3D projection and line fitting of the ground point cloudare used to reduce the computational complexity and realize the segmentation of the ground point cloud.For the collection of non-ground point clouds,SLR clustering algorithm is used to process.Ground points and non-ground points are distinguished in the vertical direction by setting the intensity feature threshold,and the scanned obstacle ground is classified.Through experimental analysis,the proposed algorithm has better mapping effect than other ground point cloud segmentation algorithms in sloping terrain.On the other hand,the intensity features processed by SLR clustering algorithm are more accurate in the coverage of X,Y and Z directions.For example,in the X direction,compared with the fast ground segmentation algorithm,the average increase is 44.0%;compared with the algorithm with grid map,the average increase is 40.1%.

关 键 词：倾斜地面 激光雷达 点云分割 栅格地图 聚类 

分 类 号：TN95[电子电信—信号与信息处理]