Reconstruction of a Botanical Tree from a 3D Point Cloud

3D models are often acquired using optical methods such as LiDAR, structured light, or automated photogrammetry. These methods produce point clouds, and the typical downstream processing pipeline consists of registration of individually scanned point clouds followed by reconstruction of a triangle mesh from the combined point cloud. In this paper we consider a specific challenge that might prevent this pipeline from producing meshes suitable for later applications. The challenge concerns reconstruction of 3D models with thin tubular features, here exemplified by a tree with a very complex crown structure, where the radii of some branches are on the same order as the sample distance. In such cases, traditional surface reconstruction methods perform poorly. We discuss how a surface can still be reconstructed from this type of data. Our procedure begins by constructing a skeleton of the object from a graph whose vertices are the input points, a surface representation is then created from the skeleton, and, finally, a triangle mesh is generated from the surface representation. We demonstrated the efficacy of our method on a tree acquired using ground-based LiDAR.