Automatic extraction of street trees' nonphotosynthetic components from MLS data

This paper aims to propose a cluster-based approach for trees' nonphotosynthetic components extraction from mobile LiDAR point clouds. The presented algorithm uses a bottom-up hierarchical clustering strategy to combine clusters belonging to nonphotosynthetic components. The combination process depends on the dissimilarity between two clusters. The measure in the proximity matrix calculation consists of a distance term using the Euclidean distance and a direction term based on the principal direction, respectively. The main contribution of this paper is to solve the optimization of cluster combination by minimizing the proposed energy function and to extract nonphotosynthetic components through a hierarchical clustering process automatically. Performance of the proposed nonphotosynthetic components extraction shows that we achieve the completeness of 94.0%, the correctness of 98.9% and the F-score of 0.96 on the experimental urban scene. Besides, we succeed to achieve promising results on the stem detection and individual tree segmentation based on the extracted nonphotosynthetic component.