Urban Building Rooftop Extraction And 3D Modeling Based On Digital Surface Model From Aerial Stereo Images

Building rooftop information and its corresponding 3D models are of great importance to the city modelling,solar energy supply,environmental simulation and disaster management,among others.A lot of research has been devoted to building rooftops extraction and 3D reconstruction from Digital Surface Model(DSM)generated from Light Detection and Ranging(LiDAR)data.Compared with photogrammetric techniques,however,LiDAR technology has the problem of higher cost of data acquisition and lower speed of data update,and thus is difficult for applying to large urban areas.Therefore,for large urban areas where no high resolution airborne LiDAR are available,automated techniques for building rooftops extraction and reconstruction from stereo aerial images are urgently needed.In this paper,by using the stereo aerial images based DSM data and green space distribution data,methods for building rooftop extraction and 3D modeling are proposed.The main research contents and results are as follows:(1)This paper proposed a method for building rooftop extraction by combining the DSM data with green space distribution data.Two filters,namely,Gaussian filter and median filter,are first used to remove noise in the DSM data.The initial seed regions of the building rooftop are then obtained by dilation reconstruction of the filtered DSM data.Since non-building rooftop regions(i.e.,vegetation and terrain regions)may exist in the initial seed regions,green space distribution data and Digital Elevation Model(DEM)data are used to remove these non-building rooftop regions.Hereafter,a region growing method is applied to the seed regions to detect building rooftop patches,and followed a region merging method to obtain the complete building rooftops by merging adjacent building rooftop patches.Finally,the final building rooftop are obtained by smoothing the rooftop boundaries by morphological processing.The proposed method was applied to the centre of Shanghai,China,a typical large urban area with various building structure types.Nine different validation areas with different land use types were selected for accuracy evaluation,and the number-based and area-based completeness are 93.52% and 96.22%,respectively.Results revealed that the proposed method can extract building rooftop planes accurately.The method proposed in our study can be extended to other large urban areas,to provide the data needs of building rooftop information for urban planning and management.(2)Based on the extracted building rooftop data and normalized Digital Surface Model(nDSM)data,this paper improves a graph-based method for 3D building rooftop modeling.Firstly,the building rooftop contours are generated,and then a set of operations including regularization,rotation and clipping,are performed to make them more regular.Subsequently,a graph-based localized contour tree method is used to represent the topological structure of building rooftop,then separates the building rooftop into different parts by topology change detection,and each single homogeneous structure of the building rooftop is obtained.The bipartite graph is constructed by using two consecutive contours,and the maximum matching result is computing of the bipartite graph for the model reconstruction.The final building rooftop 3D model is obtained by compositing all of the individual part models to a complete model.This paper used a part of Lujiazui region in Shanghai as an example to verify the method.The results revealed that complex building rooftop could be reconstructed successfully.The proposed method provides a more complete topological and geometrical description of building rooftop contours,and offers an effective solution for 3D building rooftop model reconstruction from aerial stereo images.