欢迎访问林业科学,今天是

林业科学 ›› 2016, Vol. 52 ›› Issue (3): 121-128.doi: 10.11707/j.1001-7488.20160315

• 研究简报 • 上一篇    下一篇

基于地面激光扫描数据的单木分枝结构参数自动提取

刘金鹏, 张怀清, 刘闽, 李永亮   

  1. 中国林业科学研究院资源信息研究所 北京 100091
  • 收稿日期:2015-04-14 修回日期:2015-08-04 出版日期:2016-03-25 发布日期:2016-04-08
  • 通讯作者: 张怀清
  • 基金资助:
    国家"863"计划课题(2012AA102002)。

Automatic Extraction of Individual Tree Branch Structure Parameters from Terrestrial Laser Scanning Data

Liu Jinpeng, Zhang Huaiqing, Liu Min, Li Yongliang   

  1. Research Institute of Forest Resource Information Techniques, CAF Beijing 100091
  • Received:2015-04-14 Revised:2015-08-04 Online:2016-03-25 Published:2016-04-08

摘要: [目的] 应用地面激光扫描(TLS)数据提取单木点云骨架模型,提出一种基于骨架模型的单木分枝结构及枝长、分枝角度信息自动提取方法,为单木参数化建模与森林可视化模拟提供数据基础,并丰富林业参数调查手段。[方法] 应用地面激光扫描仪器FARO,以1/4分辨率、测速244000点·s-1获取试验区3株单木的点云数据。首先,采用SkelTre算法对带有噪声的TLS数据生成骨架模型;然后,基于该骨架模型和图深度优先搜索算法提出一种自动提取分枝结构的方法,根据骨架模型邻边的拓扑关系搜索与各个分枝相连接的节点,每个节点中包含该节点主导方向、邻边个数及位置的属性,通过这些属性判定节点是否为分枝节点;由于分枝节点存在一定的位置误差,因此采用圆柱拟合法修正着枝点处分枝的延伸方向,降低应用原始骨架模型估计的分枝角度值的误差,以获取较为精确的分枝角度θ;最后,根据修正后的角度大小划分不同级别的分枝结构,通过计算分枝的骨架线段总长度,估计分枝的枝长并与野外实测值进行对比验证。[结果] 试验单木的分枝结构明确,实现了枝长与角度的自动获取,共提取出22条一级枝、43条二级枝;枝长的实测值与估计值回归方程为Y=1.003X+0.03,R2=0.998,二者的均方根误差为0.029 m;骨架模型估计的分枝角度与实测角度的回归方程为Y=0.672X+16.779,R2=0.356,二者的均方根误差为20.45°;应用圆柱拟合法修正后的分枝角度与实测角度的回归方程为Y=1.008X+0.18,R2=0.975,二者的均方根误差为3.44°。[结论] 基于骨架模型并以拟合圆柱法修正分枝角度提取单木形态结构参数的算法能够比较准确地提取单木分枝结构,枝长的提取精度较高,圆柱拟合法能够有效改善骨架模型估计角度的精度。

关键词: 地面三维激光扫描, 骨架模型, 枝长, 形态结构, 分枝角度

Abstract: [Objective] Using terrestrial laser scanning(TLS) data to extract skeleton of individual tree and developing an automatic measurement algorithm of branch structure based on skeleton model for parameterized modeling and visualization simulation of forest.[Method] We applied FARO terrestrial laser scanner, with a quarter of the resolution, speed 244 000 s-1 to obtain three trees' TLS data. Firstly, skeleton model was generated by SkelTre algorithm using TLS data with noise. Then, we developed a new method to extract branch structure with depth first search algorithm. According to the skeleton model, we searched the branch nodes connecting parent branch with child branch. Each node contained some information about this dominated direction, number of different branches and those coordinates. Due to the position error caused by skeletonization, we adopted the cylinder fitting to correct the extension direction of the branches and reduce the error of the branches' angle estimated by the skeleton model. Finally, we extracted the morphological structure based on the revised branches' angle through verification of total length of the estimated and measured branches.[Result] The experimental trees' branch structure was clear. Branch length and angle were all automatically determinated. Primary branches extracted were totally 22 and the secondary branches were 43. Regression equation of the estimated and measured length was Y=1.003X+0.03 which performed with R-squared of 0.998. The root mean square error (RMSE) between estimated length and measured length was 0.029 m. Moreover, regression equation of the branch angle estimated and measured by skeleton model was Y=0.672X+16.779, which performed with R-squared of 0.356 and RMSE of 20.45°. Nevertheless, regression equation of the branch angle estimated and measured by cylinder fitting was Y=1.008X+0.18, it performed with R-squared of 0.975 and RMSE of 3.44°.[Conclusion] The algorithm of estimation and extraction based on cylinder fitting could accurately determine branch structure on individual tree. Furtherly, the extracted branches' length performed high precision and cylinder fitting method could effectively improve accuracy of the branch angles estimated by skeleton model.

Key words: terrestrial laser scanning, skeleton model, branch length, morphological structure, branch angle

中图分类号: