基于均值高程图的城市环境三维LiDAR点云地面分割方法

Mean Elevation Map-Based Ground Segmentation Method for 3D LiDAR Point Cloud under Urban Environment

为解决智能车辆环境感知模块在地面分割过程中存在的分割精度低且耗时长的问题,提出一种应用于城市复杂环境点云的地面分割方法。首先,将三维激光雷达点云转化为均值高程图,根据高度梯度阈值将网格划分为障碍物和候选地面;然后,将候选地面的最大连通域作为参考地面,其他连通域以参考地面的高度值为基准重新划分归类;最后,通过校正梯度计算过程中产生的方法误差,进一步完善地面点和非地面点的划分,避免过分割和欠分割的产生。在城市复杂动态环境下进行实验,结果表明:与现有典型方法相比,所提方法能够实时有效地分离点云的地面区域,通过定量评估,验证了该算法在复杂城市环境下分割地面点与非地面点的准确性和鲁棒性。

To resolve problems of low accuracy and time-consuming in ground segmentation process of environment perception modules of an intelligent vehicle,the paper puts forward a new method for ground segmentation of point cloud under a complex environment. It transforms the 3D LiDAR points into mean elvation map,on which grids fall into two catogries of obstacles and candidate ground by height gradient threshold,and redivides other connected regions by height of reference ground,the maximal connected region of the candidate ground. Through the correction of errors generated in gradient calculation process,it perfects the division between ground and non-grond points in order to avoid oversegmentation and undersegmentation. The experiment results show that when compared to current methods,this new method can effectively segment ground region of point clouds under complex and dynamic urban environment. The quantitive assessment also proves this method's accurateness and robustness.