摘要
针对激光雷达不同类型点云在不同场景下存在地面点云过分割和欠分割的问题,本文提出一种能够适用于不同类型点云的地面分割算法,该算法先将原始点云栅格化,然后计算栅格单元高度差、平均高度和高度方差信息,综合三个分割指标实现地面点云准确快速分割。分别采用KITTI开源数据集和实测数据进行实验验证,结果表明本文算法针对不同类型点云在不同场景下均实现了良好的地面分割效果,平均分割准确度达到99.10%,平均耗时13.1697 ms,提升了自动驾驶汽车感知系统的鲁棒性和实时性能。
Aiming at the problem of over segmentation and under segmentation of ground point clouds in different scenes of LiDAR,a ground segmentation algorithm was proposed in this paper,which was suitable for different types of point clouds.Firstly,the original point cloud was grid,and then the grid cell height difference,average height and height variance information were calculated,and the three segmentation indexes were integrated to realize accurate and fast ground point cloud segmentation.Experiments are carried out using KITTI dataset and measured data.The results show that the algorithm achieves good ground segmentation effect for different types of point clouds in different scenarios.The average segmentation accuracy is 99.10%,and the average time consumption is 13.1697ms,which improves the robustness and real-time performance of the automatic driving vehicle perception system.
作者
邹兵
陈鹏
刘登洪
Zou Bing;Chen Peng;Liu Denghong(Chongqing Survey Institute,Chongqing 401520,China)
出处
《城市勘测》
2021年第3期112-116,共5页
Urban Geotechnical Investigation & Surveying
关键词
激光雷达
地面分割
栅格投影
高度方差
实时
LiDAR
ground extraction
raster projection
height variance
real time
