Segmentation of Forest Terrain Laser Scan Data

摘要

Modeling of realistic forest scenes is a challenge in virtual reality. This can take benefits from laser scan data acquisitions, where the segmentation of tree objects becomes an important topic. In this paper, we present a new automatic forest segmentation method. In the scan point cloud, based on normal directions, trunks are detected, leading to a correct extraction of single trees from the forest. With the trunk positions, the digital terrain model can be refined. Our technique separates the lower scene points, dedicated to trunk detection, from upper ones, dedicated to crown assignment, making the implementation easy and efficient. In addition, trunks are reconstructed, helping to construct a virtual scene and estimating tree diameter at breast height measurements (DBH). The proposed approach is implemented on real scene data, even including inclined trees, and opens an application to forestry inventories.