基于车载激光雷达的稳健地面分割方法

A robust ground segmentation method based on vehicle-mounted-Lidar

针对在不同场景下车载激光雷达地面分割存在误分割和欠分割的问题,提出了基于多区域稳健平面拟合和多条件约束的地面分割方法。首先构建多区域地面模型,为每个区域分配适当密度的点云。其次在每个区域中构建地面骨架,利用稳健平面拟合估计地面模型,通过置信度传播联合多约束条件去除非地面模型。最后利用平面拟合残差指导边界处的点云分割,进一步提升分割准确性。实验结果表明了,该方法在不同场景下综合评估指标F1能够保持在95%左右,且能适用于不同类型的激光雷达,验证了所提方法具有较强的稳健性和准确性。

Aiming at the problems of missegmentation and undersegmentation of vehicle-mounted lidar ground segmentation in different scenarios,a ground segmentation method based on multi-region robust plane fitting and multiconditional constraint is proposed.Firstly,a multi-region ground model is constructed,and point clouds with appropriate density are assigned to each region.The ground skeleton is constructed in each region,and the ground model is estimated by robust plane fitting,and then the ground model is removed by confidence propagation combined with multiple constraints.Finally,the plane fitting residuals are used to guide the segmentation of point clouds at the boundary to further improve the segmentation accuracy.The experimental results show that the comprehensive evaluation index of the proposed method can be maintained at about 95%in different scenarios,and can be applied to different types of Lidar,which verifies the robustness and accuracy of the proposed method.