Automatic Registration of Tree Point Clouds From Terrestrial LiDAR Scanning for Reconstructing the Ground Scene of Vegetated Surfaces

Abstract

Multiple scans are generally required to fully reconstruct 3-D models of botanical trees. An algorithm for the automatic registration of tree point clouds scanned from terrestrial laser scanners is proposed in this letter. The method extracts skeletons from the point cloud and conducts coarse registration automatically. It defines a distance measure between two skeleton segments and a mapping cost function between two skeletons. The coarse registration is refined using the Gauss–Newton method. Three example trees, including a Populus euphratica tree scanned in the lower reaches of the Heihe River basin, are registered using the proposed algorithm. The algorithm does not require a perfect skeleton to be extracted. No manual coarse registration is needed. The algorithm contributes to the automatic marker-free tree point cloud registration and improves field scanning efficiency by making the placement of markers unnecessary.