TreeSke: A Structural-Lossless Skeleton Extraction Method for Point Cloud Data of Canopy Woody Materials

Abstract

Background: Skeletons extracted from point clouds of woody materials present canopy structural features (e.g., the inclination angle of branches) for simulating canopy interception and understory solar radiation distribution. However, existing methods cannot easily capture structure-lossless skeletons of woody organs from tree point-cloud data. To fulfill this goal, we proposed a distance-weighted method, named the TreeSke method, to iteratively contract the point cloud of canopy woody materials to their median-axis skeleton. After heuristically searching the local point set, we pointwise-extracted the tightened weights to obtain the coarse skeleton. Then, we thinned the skeleton of woody organs and optimized it by a noise filtering and re-centering process. Results: The proposed method was verified on six simulated tree models that have reference skeletons and two field-collected datasets at the plot level. The results show that the TreeSke-extracted skeletons were with higher location accuracy than using the other two tested methods. The mean offset distance, the RMSE of the offset distance, and the Hausdorff distance between the extracted and reference skeletons were less than 0.04 m, less than 0.06 m, and less than 0.11 m, respectively. Conclusions: The extracted skeletons by the TreeSke method could nearly integrally depict the structural features of woody materials. The proposed method showed robust for noisy points and outliers, while missing data would reduce the skeleton integrity and cause deformation errors. From the extracted skeletons, users can extract the inclination-angle features of canopy woody materials, which are useful for simulating most eco-hydrological processes.

Tree Information Modeling: A Data Exchange Platform for Tree Design and Management

“…At this step, improvements can be made regarding computing time and tolerance to point clouds in low quality (see Section 3.1). (2) To abstract the topological skeleton of branches out of the point cloud, Cornea [136] compared multiple automatic skeletonization methods; L1-medial skeleton [137] is efficient on point cloud that is not over complex containing too large an amount of points; [138] developed an approach to restore a speculative skeleton without segmenting point clouds into branches and leaves; Wu et al [139] then achieved an accurate medianaxis skeleton abstraction based on the foliage-woody separation by convolutional neural networks [140]; Liu et al [141] developed a neural network to reconstruct tree geometry out of a point cloud robust to noise, outliers and incompleteness; besides, voxel thinning is able to preserve the precise topological structure of tree branches while estimating approximate diameters of branches during the thinning process [142]. (3) After skeletonization, pipelines can be generated by cylinder fitting or calculating the average distance from the trunk surface to the skeleton on perpendicular planes.…”