联合UAV-LiDAR和HMLS技术的森林样地点云数据融合  

Point cloud data fusion of forest plots based on UAV-LiDAR and HMLS technologies

作　　者：王楚虹 刘浩然 钟浩 林文树[1] WANG Chuhong;LIU Haoran;ZHONG Hao;LIN Wenshu(College of Engineering and Technology,Northeast Forestry University,Harbin 150040,Heilongjiang,China)

机构地区：[1]东北林业大学工程技术学院,黑龙江哈尔滨150040

出　　处：《中南林业科技大学学报》2022年第3期26-38,共13页Journal of Central South University of Forestry & Technology

基　　金：国家自然科学基金项目(31971574);黑龙江省自然科学基金联合引导项目(LH2020C049);中央高校基本科研业务费专项资金项目(2572019BL03)。

摘　　要：【目的】基于单一平台遥感数据提取的森林结构参数信息不全面,因此融合多平台遥感数据已成为遥感技术在林业应用中的发展趋势。本研究针对手持移动激光雷达和无人机激光雷达2种不同平台的数据,提出一种基于Delaunay三角网和迭代最近点(ICP)算法的点云数据融合方法,通过融合获得完整的树木点云数据。【方法】选取哈尔滨城市林业示范基地中樟子松和蒙古栎2块人工林作为研究样地,利用手持移动激光雷达和无人机激光雷达获取样地中树木点云数据,然后分别提取2种不同平台点云数据的树木位置点,使用2组位置点构建2个Delaunay三角网来搜索树木位置目标点与其对应点,以此实现点云的粗配准,并利用模拟退火算法优化粗配准过程。在此基础上,通过选取点云重合度高的部分树冠点云,利用ICP算法进行点云数据的精配准,从而实现2种平台森林样地点云数据的精确融合。【结果】粗配准和精配准中分别设置的线段差阈值和高度区域范围这2个参数对点云数据的配准有较大的影响,而粗配准中迭代次数参数的设置对融合结果的影响较小。在样地内部随机选取区域,将区域内配准后的2种平台点云的树木位置投影点坐标进行对比,得到樟子松和蒙古栎的投影点平均坐标偏移距离分别为0.19和0.25 m。根据采集数据时设置的标志物计算偏移误差,樟子松样地和蒙古栎样地融合结果的均方误差分别为0.0512和0.0802,樟子松样地点云数据的融合精度高于蒙古栎样地。【结论】本研究提出一种基于树木位置的不同激光雷达平台点云数据无标识融合方法,融合精度较高,空中与地面获取的点云数据实现了相互补充,可为森林结构参数的精确提取及树木三维模型构建提供数据支撑,从而推动激光雷达数据在森林资源调查等方面更加广泛地应用。【Objective】As the forestry information extracted from remote sensing data based on a single platform is incomplete,integrating multi-platform remote sensing data has become the development trend of remote sensing technology in forestry applications.This research focused on the data from two different platforms including handheld mobile laser scanning(HMLS)and UAV LiDAR,and a point cloud data fusion method based on Delaunay triangulation and Iterative Closest-Point algorithm(ICP)was proposed and then the tree point cloud data was completely obtained.【Method】Two plots of Pinus sylvestris and Quercus mongolica in Harbin Urban Forestry Demonstration Base were chosen as the research plots and the tree point cloud data were obtained by using HMLS and UAV LiDAR.Then two sets of tree location points were extracted from the point cloud data of the two different platforms,and two Delaunay triangulation networks were constructed to realize the rough registration of the point cloud.A simulated annealing optimization algorithm was used to optimize the rough registration process.Based on the rough registration,by selecting parts of canopy point clouds with high point cloud coincidence,ICP algorithm was used to carry out precise registration of the point cloud data,to achieve fine fusion of forest plot point cloud data of the two platforms.【Results】The two parameters,the line segment difference threshold and the height area range respectively set in the rough registration and the precise registration had a great impact on the registration of the point cloud data.The number of iterations set in the rough registration had little influence on the fusion results.By comparing the coordinates of tree position projection points of the two platform point clouds after registration randomly selected inside the plots,the offset error was calculated according to the markers set during data collection.The average coordinate offset distances of projection points of the Pinus sylvestris plot and the Quercus mongolica plot were 0.1

关 键 词：手持移动激光雷达 无人机激光雷达 点云融合 DELAUNAY三角网 ICP算法 

分 类 号：S771.8[农业科学—森林工程]