A Semantic Segmentation Algorithm of Ancient Building's Point Cloud Data

Abstract

In view of the difficulty of wooden elements segmentation of three-dimensional laser scanning point cloud data of ancient building, a new algorithm for effective and accurate segmentation of ancient building's point clouds is presented. Firstly, point cloud denoising is implemented. Then a section intersecting with wooden columns and being perpendicular to the vertical axis of the three-dimensional coordinate system is created based on structural characteristics of ancient buildings. At the same time, point clouds on the section are extracted and divided into several parts which are corresponding to different wooden columns based on point cloud Euclidean clustering. Parameters of wooden columns are estimated and used to compute the Rodrigo rotation matrix in order to implement automatic spatial transformation of ancient building's point clouds. By this processing, vertical axis of the three-dimensional coordinate system will be strictly perpendicular to the ground. And then point clouds of wooden columns are extracted by model fitting method. Other wooden elements (such as beams, tiebeams etc.) are segmented out by bounding box created by the information of structure and size of ancient building. In order to prove the robustness and feasibility of the algorithm, the point clouds of pavilions are selected for the experiment in this study. The result shows that the algorithm is feasible and robust. This research can provide supports of theory and methods for deeply research of automatic segmentation of ancient building's point cloud data.