联合空-车LiDAR点云的城市树木精细提取及重建方法研究

In view of deficiencies in points cloud occlusion from single LiDAR platform, fine extraction of tree structure and tree modeling, the project aim at studying precise extraction and reconstruction of urban trees from airborne-vehicle combined point clouds. On the basis of analysis of urban road features, a registration method between airborne and vehicle LiDAR points cloud by road network nodes and road centerlines was developed to offer basic data for airborne-vehicle LiDAR combined data processing. In consideration of irregularity of LiDAR points cloud, morphological characteristics of urban trees were explored to differentiate other objects. Furthermore, a segmentation method for tree points cloud by analysis of super-voxel characteristics was developed to extract tree points. According to the characteristics of tree canopy structure, top points of tree crown were detected by analysis of horizontal grid data and circular grid data, and then an individual tree extraction method based on crown top points was realized by an improved watershed algorithm; At last, skeleton was rapid detect from single tree points and a method of tree modeling based on skeleton reconstruction was developed to accomplish reconstruct tree model. The method of registration of airborne-vehicle LiDAR point clouds will be innovated and precise extraction and reconstruction of urban trees will be developed. The research achievements will provide technical foundation for information extraction and further application of urban trees and meet the professional needs for urban environmental studies.

针对单一LiDAR平台获取树木点云存在遮挡，现有方法难以提取树木精细结构以及树木建模效率低等问题，本项目拟开展空-车LiDAR点云联合的城市树木精细提取及重建研究。在分析城市道路特征的基础上，研究基于路网节点及道路中心线的空-车LiDAR点云配准方法，为空-车LiDAR点云联合处理奠定数据基础；深入分析点云分布特点，探索城市树木区别于其他地物的形态特征，研究基于超体元特征分析的林区点云分割方法，提取城市树木点；根据树木冠层结构特点，分析联合水平格网及环状格网的树冠顶点检测方法，实现基于分水岭分割的单株树木提取；根据单株树木点云探索骨架线快速检测方法，并研究基于骨架线重构的树木建模方法，构建单株树木模型。研究成果有望在空-车LiDAR点云联合及城市树木精细提取与重建的方法上有所突破，并为城市树木信息提取及深化应用奠定基础，进一步满足城市环境研究的专业需求。

为了克服单一LiDAR平台获取的树木点云存在遮挡，解决树木精细结构提取困难以及模型重建效率低等问题，本项目开展了联合机载与车载激光点云的城市树木提取及模型重建方法研究。项目首先研究了基于道路特征及ICP算法的机载点云与车载点云配准方法，该方法利用机载与车载场景的公共要素道路作为基元进行点云粗配准，然后利用基于KD树的ICP算法实现机载与车载点云精配准。针对树木点云识别，项目研究了点密度自适应的地面激光扫描点云分类方法，该方法能根据目标距离测站越远点密度越稀疏的特点，利用极坐标网格法自适应获取邻域点，从而提高目标点的分类精度。同时项目研究了基于超体素特征分析的行道树提取方法，该方法在分析非地面点云维度特征的基础上，利用维度特征聚类方法得到超体素，然后通过超体素特征分析以及树木语义约束实现行道树点云的提取。为了分割出单株树木点云，项目以树干位置作为初值，研究了一种基于位置初值的分水岭分割方法，并利用距离竞争标记策略提高单株树木点云的分割效果。针对树木三维重建，本项目提出一种枝杈分级的树木模型重建方法，该方法利用Dijkstra算法获取树木的初始骨架，在此基础上通过枝杈分级以及节点优化处理得到树木骨架线，并利用树干圆台模拟及树叶纹理映射完成树木三维建模。研究成果可为城市树木信息提取及深化应用奠定基础，为城市环境研究的专业需求提供技术支撑。