一种基于地面特征与树木位置关系的无人机和地基LiDAR点云配准方法

A point cloud registration method for UAV and TLS LiDAR based on ground features and the relationship of tree positions

【目的】无人机激光雷达与地基激光雷达的工作方式不同,导致无人机点云缺乏林内信息,地基点云缺乏林冠信息,单一平台的LiDAR点云难以完整描述森林三维垂直结构,将这两者点云融合有利于消除各自的扫描盲区,估测更为准确的森林结构参数。基于此,提出了一种基于地面特征与树木位置关系的无标识自动化配准方法。【方法】选取哈尔滨城市林业示范基地内的蒙古栎和樟子松作为研究对象,采用大疆禅思L1激光雷达设备与FARO Focus3D X330三维激光扫描仪分别获取样地内的无人机和地基LiDAR点云。首先,利用改进的渐进式加密三角网滤波算法分别从无人机点云和地基点云中提取地面点云,基于两者相似的快速点特征直方图(FPFH)特征,使用随机采样一致性算法得到初始配准参数,完成初始配准。然后,从初始配准后的无人机点云和地基点云中提取相同高度处的树干点云的水平投影位置作为配准基元,分别构建不规则三角网,并基于三角形的角度相似性原理寻找同名三角形对。最后,使用奇异值分解法得到旋转平移参数,从而完成精细配准。【结果】蒙古栎样地内对应树木水平偏移距离的平均值为0.173 m,樟子松样地内对应树木水平偏移距离的平均值为0.283 m,2个样地的树木点云均取得了较高的配准精度。【结论】提出的点云配准方法有效实现了林区无人机点云数据和地基点云数据的配准,二者的融合可为快速完整地获取林木构型信息提供数据基础,从而推动多源激光雷达技术联合应用于林木三维重建和森林资源精细调查等方面。

【Objective】UAV LiDAR and TLS LiDAR work in different ways,resulting in the lack of in-forest information in the UAV point cloud and the lack of forest canopy information in the TLS point cloud,and it is difficult for the LiDAR point cloud of a single platform to completely describe the three-dimensional vertical structure of the forest,and the fusion of the two is beneficial for eliminating the scanning blindness of each,and for estimating the more accurate parameters of the forest structure.Based on this,a markerless automated alignment method based on the relationship between ground features and tree positions is proposed.【Method】Quercus mongolica and Pinus sylvestris in Harbin Urban Forestry Demonstration Base were selected as the research objects,and DJI Zenmuse L1 LiDAR equipment and Faro Focus3D X3303D laser scanner were used to obtain the UAV and TLS LiDAR point clouds in the sample plots,respectively.Firstly,the improved progressive encrypted triangular mesh filtering algorithm was utilized to extract the ground point cloud from the UAV point cloud and the TLS point cloud respectively,and based on the similar fast point feature histogram features of the two,the random sampling consistency algorithm was used to obtain the initial alignment parameters to complete the initial alignment.Then,the horizontal projection positions of the tree trunk point clouds at the same altitude were extracted from the initially aligned UAV point cloud and TLS point cloud as the alignment primitives to construct the irregular triangular mesh,and search for the pairs of triangles with the same name based on the principle of angular similarity of triangles.Finally,the singular value decomposition method was used to obtain the rotational translation parameters to complete the fine alignment.【Result】The average horizontal offset distance of the corresponding trees in Q.mongolica sample site was 0.173 m,and the average horizontal offset distance of the corresponding trees in P.sylvestris sample site was 0.283 m,and the point clouds of the two sample sites had achieved high alignment accuracy.【Conclusion】The point cloud alignment method proposed in this study effectively realizes the alignment of UAV point cloud data and TLS point cloud data in forest areas,and the fusion of the two provides a data basis for rapid and complete acquisition of forest tree configuration information,thus promoting the joint application of multi-source LiDAR technology in forest tree 3D reconstruction and fine forest resource survey.