DENSITY ADAPTIVE PLANE SEGMENTATION FROM LONG-RANGE TERRESTRIAL LASER SCANNING DATA

Plane segmentation is a commonly used approach that extracts detailed building information from point cloud. However, the change of density is not obvious for mainstream research data. The buildings with different densities are difficult to be classified especially for TLS data of which scanning distance exceeds 500m, while neighborhood radius is the key factor to solve this problem. In this article, an approach for density adaptive plane segmentation is presented. Firstly, compared with methods based on fixed radius range, dynamic neighborhood radius is selected before plane segmentation to ensure that the objects with different densities can be identified and the dimensionality feature of each point can be computed. Then, an improved growing rule based on dimensionality feature is applied to segment the buildings into planes. The experimental results show that the proposed method can efficiently extract planes from long-range TLS data, the precision reaches 95%, the recall reaches 92%.