Height adaptive LiDAR segmentation for building extraction and roof reconstruction

This thesis presents a new LiDAR (Light Detection And Ranging) data segmentation method for automatic building detection and roof plane extraction. First, it uses a height threshold, based on the digital elevation model (DEM), to divide the LiDAR point cloud into ‘ground’ and ‘non-ground’ points. The ‘non-ground’ LiDAR points are classified as coplanar and non-coplanar based on a coplanarity analysis. Then, starting from the maximum LiDAR height, and decreasing the height at each iteration, it looks for points to form planar roof segments. At each height level, it clusters the points based on the 2D Euclidean distance for each cluster in order to find straight lines. The coplanar point nearest to the middle of each line is selected as a seed point and the plane is grown in a region growing fashion until no new points can be added. Finally, a rule-based procedure is followed to remove planar segments on trees. Experimental results on seven data sets indicate that the proposed method offers high building detection and roof plane extraction rates when compared to existing promising methods. Next, an automatic 3D building roof reconstruction method is proposed based on the outcome of the proposed segmentation method. The method commences with the building boundary points to extract the 3D internal and external corner points on each building. Then, it connects these points to get a 3D roof model of a building. To generate a complete 3D view of a building, the ground heights of the external corner points are considered from the DEM to generate the floor of a building. By connecting the external corner points of ground and non-ground heights, the approximated building walls are generated. The method is examined with three different data sets consisting of several types of complex roof structures. Experimental results show that the proposed reconstruction method improves the correctness and reduces the over and under-segmented roof planes when compared to existing methods.