一种多阶段渐进式复原提取点云中杆状物的方法

An extraction method for pole-like objects from point cloud via multi-stage progressive data restoration

针对车载激光点云杆状物提取中,出现的点云粘连和部分点云缺失问题,该文提出了一种采用多阶段渐进式复原思路提取点云中杆状物的方法。根据点云投影前后数量一致性进行地面滤波,实现地面点与非地面点的分离;利用一种迭代穿透性因子滤波,将非地面点云分割为植被点云和非植被点云;在此基础上,使用圆柱性和垂直性检测杆状点云,并引入豪斯多夫距离合并点云缺失杆状物;利用提取的树干寻找并合并对应的树冠点云,复原完整树木。实现对车载激光点云中人工杆状物和行道树的完整提取。基于城区和郊区车载点云数据的实验结果表明:该文所提方法能适应点云粘连、点云缺失等情况,能够初步改善点云粘连和部分点云缺失情况,也能适应不同的复杂道路环境。

In response to the issues of point cloud adhesion and partial point cloud loss in the extraction of pole-like objects from vehicle-mounted LiDAR point clouds,this paper proposes a method that utilizes a multi-stage progressive restoration approach to extract pole-like objects from the point cloud.Firstly,Based on the consistency of point cloud projection in terms of the pre-and post-projection point count,a ground filtering algorithm is applied to achieve the separation between ground and non-ground points.Furthermore,an iterative penetration factor filtering technique is utilized to segment the non-ground point cloud into vegetation and non-vegetation point clouds.Building upon this,a cylindrical and vertical detection method is employed to identify pole-like point clouds,and the Hausdorff distance is introduced to merge fragmented point clouds.Finally,the extracted tree trunks are used to locate and merge corresponding tree crown point clouds,resulting in the reconstruction of complete trees.The method achieves comprehensive extraction of artificial poles and roadside trees from the vehicle-mounted LiDAR point clouds.Experimental results based on vehicle-mounted point cloud data in urban and suburban areas demonstrate that the proposed method is capable of addressing point cloud adhesion,point cloud loss,and can preliminarily improve the issues of point cloud adhesion and partial point cloud loss.Moreover,it is adaptable to different complex road environments.