Segmentation And Boundary Extraction Of LIDAR Data

With the development of modern surveying technology, 3D Laser Scanning technology has an increasingly important role in ancient architecture, reverse engineering and so on. But the point cloud data got from the 3D Laser Scanning is easily to be millions of data points. In this way, in the field of the protection of ancient architecture, most of the scale of ancient architectures is magnificence, and structure is complex, such as the forbidden city, tens of millions of point cloud data will be got. To show, transmission, storage or query the whole point cloud data model or the model of triangulation, will bring challenges to the existing computer technology. At the same time, a mass of module of the point cloud bring many kinds of trouble to the protection, management of ancient architecture and the rebuild of the 3D geometry model and analysis its configuration. So, if we can extract the plane information from the point could through region segmentation, and use the plane to express all the triangulations that are in the same plane, through the method, the large amount of data will be compressed. Meanwhile, boundary extraction through region segmentation and saving the line structure, the amount of data reduce a lot. The technology in the fields of reverse engineering, heritage preservation and archiving has an important role.The paper first build the topological relations for the point data through 3-Dgrids,then in every grid, the point cloud is reduced according to the threshold of distance and angle. Its purpose is to improve the efficiency of the later data processing of the point cloud, without changing the geometry characteristics of the point cloud. After the pretreatment, we rebuild the 3-D grids for the point cloud, two kinds of segmentation methods are proposed in the paper, the first is based on the idea put forward by I·Stamos and P·K·Allen in 2000 to realize the segmentation of plane; the other is by coordinate transformation method to get the differential geometry information such as normal vector, curvature to realize the segmentation of point cloud based on multivariate including normal vector, curvature.Saving the result of segmentation. Then through the method of Octree cube to extract the boundary of the saving point cloud, and use the Octree indexing method to establish the border point cloud index, we send the index of the array elements and the number of points to the graphics function of DirectX to achieve the extraction of the boundary of point cloud data. And we use the forward-by-point method to achieve the extraction of feature points, as well as smoothing the boundary line.According to the related researches, a 3D application system is developed in C# 2005, DirectX graphic library to validate the feasibility and the accuracy of the algorithms. The points cloud obtained by LEICA HDS4500 and LEICA HDS3000 are used in the experiments.Different kinds of methods are used to solve the display and geometric transformation of point cloud data in the 3D environment. In the experimental system of 3D operation and display environment, we rebuild the 3D-grid on the point cloud, and after that, we realize the reduction, segmentation and boundary extraction for the data, to prove the feasibility and practicality of the experimental system.