Hybrid Segmentation of Urban Trees from Airborne LiDAR Data Using DBSCAN and Watershed Algorithms

Urban trees have an essential role to improve a quality of environment, to regulate microclimate and to contribute to durable development of cities. However, tree detection from airborne LiDAR data still complex, especially in urban environment where the crowns frequently overlap. In this article, we explored a hybrid approach unsupervised for individual urban trees segmentation, based on the combined to Watershed algorithm and DBSCAN (Density-Based Spatial Clustering of Applications with Noise) method. This study reposed in airborne LiDAR data with density between 10- 20 pts/m2, delivered by the National Institute of Geographic and Forest Information (IGN) covering an urban area in the Essonne department in France. Two processing approaches are evaluated: (i) applying Watershed to a Canopy Height Model, followed by refinement with DBSCAN, and (ii) identifying the tree areas with DBSCAN first, then detecting individual trees within each cluster using Watershed. The results demonstrate that the DBSCAN followed by Watershed approach achieves more robust segmentation in dense vegetation areas, with a precision of 0.94, recall of 0.92 and Fl-score of 0.93, outperforming the Watershed followed by DBSCAN approach, which obtained 0.91, 0.81 and 0.86, respectively. This study highlights the complementarity of morphological and density-based approaches, as well as the importance of pre-processing steps. The results confirm the potential of unsupervised hybrid approaches as an effective and interpretable solution for urban tree segmentation from airborne LiDAR data.