Three-dimensional tree monitoring in urban cities using automatic tree detection method with mobile LiDAR data

抄録

Trees in urban spaces play important roles such as cooling heat islands, providing shade and biodiversity, and purifying the air. Monitoring tree structures such as tree trunk diameter, height, and woody dry weight is crucial for tree management. In this study, light detection and ranging (LiDAR) measurements were performed to monitor trees in urban spaces in lieu of manual measurement. Velodyne LiDAR and a mobile mapping system (MMS) were used for the tree measurement. Then, a method for automatic tree detection using machine learning and deep learning technique for 3D LiDAR point clouds was proposed. The proposed method can identify trees even when there are many non-tree objects in the environment. First, each object was segmented using 3D point cloud processing, and the segmented objects are projected onto 2D images. The projected images are then classified into tree or non-tree objects based on structural features using a support vector machine (SVM). In this step, a generative adversarial network (GAN) was utilized to augment the training images. Subsequently, 3D point clouds of the tree identification results are generated after the 2D-based classification. The classification accuracy of tree and non-tree 2D images after segmentation exceeds 95.0%, a significantly high degree of accuracy. Finally, tree structural information (e.g., woody dry weight) is calculated using a regression curve expression proposed in a prior study. This method enables an automatic monitoring of tree structures—for example, in urban forests, parks, and roads—using LiDAR.