Study On Building Detection And Roof Outline Extraction Based On Airborne LiDAR Point Cloud Data

Buildings are the most primary element in cities. As digital city becomes a hot topic recently,3D city models especially buildings are in great demand. Aerial photogrammetry is one of the traditional methods in 3D city models collection, which is very mature. However, there are some problems in 3D models reconstruction using aerial photogrammetry, such as the loss of geometry information of surface features, the low height precision and so on. Most of these problems are caused by image data itself, which is two-dimensional. Moreover, the selection of ground control points and stereo image matching can also reduce reconstruction precision. Airborne LiDAR, a new booming active sensor which is comprised of Positioning and Orientation System(POS) and laser range finder, can directly acquire discrete dense 3D point cloud data of surface features. This technology provides great convenience for digital city construction, especially building reconstruction.Based on the analysis about existing algorithms of building detection and roof outline extraction, the details of this dissertation are as follows:(1)The filtering process of LiDAR cloud points. This dissertation uses a modified moving window filtering method combined with region growing strategy to remove sagging surface features such as scarps. Result shows that this method works well in flat area and avoids the interference of surface features those has similar step characteristic like buildings.(2)Building detection algorithm. The dissertation uses a method that based on curvature change, coplanar judgment, elevation difference threshold and region growing. This novel method combines building inner information with roof outline information. The experiment result shows this algorithm can also apply to curved surface roofs.(3)A novel roof outline extraction algorithm is proposed in this dissertation. According to the principle that there is no point out of the line segment between two neighbor edge points, detect all edge points of each roof successively. Fit outlines by least square methods, and then do regularization and extension to outlines. The experiment result shows this algorithm can extract building roof outlines fast and precisely.