Segmenting Individual Trees From Terrestrial LiDAR Data Using Tree Branch Directivity

Abstract

Over the last decade, a number of techniques for individual tree segmentation has been developed for terrestrial laser scanning data. The superpoint segmentation algorithm based on point cloud has been widely used in individual tree segmentation because of its high efficiency and numerous geometric features. However, this algorithm is generally developed for specific tree species and forest types, limiting its universality and performance for different forest types. To handle this problem, a new method based on the topology of tree branches for individual tree segmentation was proposed. Focusing on the general topological structure of trees, the proposed method iteratively assigns each branch based on its directivity to its upper branch at the superpoint level. The proposed method was tested compared with the original superpoint method and an ecological method in six sample plots with different forest conditions. In such plots, the proposed method achieved anticipated performance with an average accuracy of 40% improvements compared with the other two methods, especially in complicated forest conditions. Experimental results also showed an improved average accuracy of 70% compared with the original superpoint method at the point level. This proposed method can effectively expand the universality of the superpoint method to further advance ecological and forest research.

Keywords

• Directivity
• individual tree segmentation
• point cloud
• superpoint
• terrestrial laser scanning (TLS)