SkelTre - Fast Skeletonisation for Imperfect Point Cloud Data of Botanic Trees

Abstract

Terrestrial laser scanners capture 3D geometry as a point cloud. This paper reports on a new algorithm aiming at the skeletonisation of a laser scanner point cloud, representing a botanical tree without leafs. The resulting skeleton can subsequently be applied to obtain tree parameters like length and diameter of branches for botanic applications. Scanner-produced point cloud data are not only subject to noise, but also to undersampling and varying point densities, making it challenging to extract a topologically correct skeleton. The skeletonisation algorithm proposed in this paper consists of three steps: (i) extraction of a graph from an octree organization, (ii) reduction of the graph to the skeleton and (iii) embedding of the skeleton into the point cloud. The results are validated on laser scanner point clouds representing botanic trees. On a reference tree, the mean and maximal distance of the point cloud points to the skeleton could be reduced from 1.8 to 1.5 cm for the mean and from 15.6 to 10.5 cm for the maximum, compared to results from a previously developed method.