基于三维激光雷达的复杂场景中地面分割方法

Ground Segmentation Method in Complex Scenes Based on Three-dimensional Lidar

针对现有地面分割方法在复杂场景存在鲁棒性差和现有传统方法依赖阈值的问题,提出了一种合并式地面分割算法。首先根据车辆行驶实际情况进行预处理;然后对基于射线特征和基于平面模型拟合的方法进行改进与分阶段式融合,第一阶段中加入最大阈值的思想实现粗分割,第二阶段中加入结合道路分布特征和激光雷达线束分布特征的多区域划分思想实现细分割;最后对非地面点进行杂点过滤。实验结果表明,所提方法在复杂场景进行地面点分割的平均精确率有显著提升,平均召回率取得较高水平且无明显波动,单帧精确率和召回率均在0.85以上,具备优良的鲁棒性,且无需根据路面坡度的实际情况调整阈值,具有普遍的适用性与实际意义。

The existing ground segmentation methods have poor robustness in complex scenes and the existing traditional methods rely on thresholds. Hence, a combined ground segmentation algorithm is proposed to address these problems. First, preprocessing is performed according to the actual driving situation of the vehicle. The methods, based on ray features and plane model fitting, are then improved and merged in stages. In the first stage,the idea of adding the maximum threshold is added to achieve coarse segmentation, and in the second one, the idea of multiregion division combining road distribution characteristics and lidar beam distribution characteristics is added to achieve fine segmentation. Finally, nonground points are filtered. The experimental results show that the average precision of the method proposed in this paper for ground point segmentation in complex scenes has been significantly improved, and the average recall rate has reached a high level with no significant fluctuations. The single-frame precision and recall rate are both greater than 0. 85, indicating excellent robustness. The proposed algorithm has no need to adjust the threshold based on the actual situation of the road slope, which has universal applicability and practical significance.