Identification of Tree-Shaped Structure using 3D Reconstruction and Density-Based Clustering

Abstract

In 3D reconstruction, there are problems when reconstructing irregularly and complexly overlapping objects such as trees. This research aims to reconstruct three-dimensional linear structures such as trees, bent reinforcing bars, and tangled wire waste. We propose a method to identify the tree-shaped structure of a linear three-dimensional structure by combining 3D reconstructed point cloud data using moving images and voxel-based denoising with density-based clustering. We report the results of the 3D reconstruction of the tree structure of the entire deciduous fruit tree using the proposed method and demonstrate the effectiveness of the method. In addition, the branching structure of the linear 3D structure could be represented from the tree diagram of voxels.