| dc.description.abstract | Lidar is widely used for the applications of the Cyber city construction and 3D reconstruction of buildings, because it is capable of fast and accurate data acquisition from distant objects. Traditionally, the Triangulated Irregular Network (TIN) is used to create 3D models from a point cloud. Although the generated models are realistic, TIN creates a huge number of triangles and causes inefficient computation and data storage. Furthermore, one of the most significant properties of buildings is that they are mostly composed of planes and have more regular shapes than other types of objects. Therefore, this research aims to integrate octree structure, which is effective in extracting planes in a point cloud, and grid adaptation model, which is effective for reconstructing detailed parts of an object, for the automated generation of 3D models. In the first part, the planes are extracted, if existed, based on the points in a cell. If a plane cannot be found, the cell is split using an octree structure. It is found that considering initial cell sizes, point distribution in the cell, boundaries of the points, and adjustment for the point distance will improve the correctness of the 3d models. In the second part, to avoid the errors caused by small cell sizes, the splitting stops when the size of a cell is smaller than a threshold and the points are then transferred to the grid adaption models. The main consideration in the grid adaptation model is the selection of the representative point based on the relationship between adjacent cells. The planes found in the both parts are combined for the final model. Test results show that the octree structure and the grid adaptation model are successfully combined and can be used for automated 3D reconstruction for buildings.
| en_US |