An automated approach for wood-leaf separation from terrestrial LIDAR point clouds using the density based clustering algorithm DBSCAN

The terrestrial light detection and ranging (LiDAR) technique has been recently used to provide 3D structural information of forest canopy at the individual tree level. However, the operational use of Terrestrial Laser Scanner (TLS) for canopy characterization of broadleaf non-deciduous forests needs further investigations. The estimation of wood volume, above-ground woody biomass, tree canopy characteristics and leaf area index often requires separation of photosynthetically active material and non-photosynthetically active material. This article describes an automated wood-leaves separation method, based on spatial geometric information of TLS point clouds, for broad leaved non-deciduous trees. Scans of seven individuals of Quercus suber L. trees were acquired by using the TLS phase-based Leica HDS6100. Point clouds were partitioned in cubic volumes (voxels) that were used as input to generate clusters through the point density algorithm DBSCAN. The clustering process led to the identification of wood and non-wood voxels. A specific automatic routine was written to process data from the point clouds to the visualization of clustering results. The analysis of results showed good performance for this approach, with the overall accuracy in classifying wood components of trees ranging from 95% to 97%. The largest accuracies were observed for branches larger than 5 cm in diameter whereas the accuracy of classification dropped, as expected, for branches with diameter lower than 3 cm. The results suggest that the proposed method can be conveniently used to extract woody components from point clouds of broad leaved non-deciduous trees.