| At present,3D model of buildings has been widely used in the city planning,homeland security and disaster management and other aspects.Building 3D reconstruction has become a research hotspot in photogrammetry,remote sensing and computer vision.Thus,this paper focus on building extraction and structured 3D reconstruction of building model.The main work and innovations are listed as follows.1.Building 3D reconstruction method based on image data,point cloud data and multi-source data have been analyzed systematically.Then,the main problems in this research area are summarized,and it laid the foundation for the following research in building extraction and 3D reconstruction.2.For point cloud filtering in urban areas,a point cloud data filtering method based on the guidance of the block surface gradient has been proposed.The block size is determined by the density of point cloud,and mask area is constructed based on gradient between blocks.Then,mask area and non-mask area are filtered respectively,and the ground points and non-ground points are separated.This method can resolve the problem that large buildings are hard to be filtered with small block size.3.A building point cloud detection method is presented based on multiple features fusion.The surfaces of point cloud in building area and vegetation area have different rolling condition,and it can be reflected by the direction of normal.After analyzing the space characteristic of point cloud and relative images features,fusion of various spatial characteristics and image features are accomplished through the Support Vector Machine,and the non-ground points are classified as either buildings or vegetation.Then the detection result is optimized through neighborhood characteristics,and other constraints such as height and area are used for removing non-building areas further.At last,morphological method is adopted for labeling building area.Comparative experiments show that this method has high accuracy and geometric precision.4.The traditional RANSAC algorithm in building roof point cloud segmentation is optimized.By constructing the normal criterion using local surface normal constraint,coplanar roof point cloud is decomposed by a radius constrained method of point cloud spatial clustering,thus suppressing the generation of "false roof plane".Then,a local sampling strategy is used to reduce the iteration numbers of the algorithm,and reliable segmentation results of building roof point cloud can be obtained.5.A structured 3D reconstruction method of building roof is designed based on perceptual grouping.A series of perceptual grouping principles are used for constructing wall space equation,such as parallel and vertical,the same regional and adjacent,proximity,similarity,continuity and closed rules.Meanwhile,both the relationship between the roof surfaces and the relationship between the roof and walls is analyzed.Finally,regular building structure line is obtained and 3D building models are reconstructed.Experiments have confirmed the effectiveness and veracity of this method,which can accomplish the reconstruction of both simple buildings and complex multilevel buildings.6.Considering rich and disorder edge features in high resolution image,the building model optimizing scheme is designed.The building roof contour line features are extracted based on the initial building roof model.Then the corners of building in image can be acquired by these line features,and the initial building model can be optimized by the combination of collinearity equations and roof plane equations.7.A 3D model reconstruction system for buildings is designed and achieved.Through analyzing the components and functions,the design and development of the system is accomplished.It can support loading multiple format of point cloud and image data,and realize multi-angle stereo viewing for point cloud and model data.It can also detect buildings and reconstruct regional buildings in batch,which makes the system more practical. |
PDF Full Text Request