Research On 3D Model Reconstruction From Airborne LiDAR Point Cloud Data

As a dominant kind of manmade object in city scenes,the automatic reconstruction of buildings has been a focused and difficult research problem in the field of photogrammetry and remote sensing and computer vision.With the abilities of directly and quickly obtaining dense high precision large-scale region point cloud,airborne LiDAR(Light Ditection And Ranging)technology can provide reliable data for building 3D model reconstruction.However,due to the various factors such as irregular distribution of airborne LiDAR point cloud data and complex building spatial structure in real scene,automatic reconstruction of building 3D model based on airborne LiDAR point cloud data is still a challenging task.Under this background,the dissertation focuses on the method of building 3D model automatic reconstruction based on airborne LiDAR point cloud data.The major work and innovations of the dissertation are listed as follow:1.After introduction the background and significance of building 3D model reconstruction,methods of building 3D model reconstruction based on different data resources are concluded,reconstruction methods based on point cloud are mostly analyzed,commonly used building structure knowledge is summarized,multi-types building structure representation methods are comparatively analyzed.2.Method of extracting building point cloud based on building boundary points region growing is proposed.The method extracts initial building boundary points according to the geometric characteristics of buildings,distribution of point cloud is then calculated by constructing the local covariance matrix to eliminate false building boundary points,multi-seed points region growing is applied to extract building point cloud based on building boundary points cluster result by classic density based clustering algorithm.Comparative analysis with existing methods is made through two sets of experiments,and effectiveness of the method is verified.3.Aiming at the poor adaptation and low extraction precision of existing roof extraction methods for complex building,an automatic building roofs extraction algorithm using neighborhood information of point clouds is presented.Feature histogram is constructed based on point cloud curvature,and reliable seed points are selected after the histogram construction.Initial roof surfaces are extracted quickly and precisely by the proposed local normal vector distribution density-based spatial clustering of applications with noise(LNVD-DBSCAN).Roofs competition problem is solved effectively by the poll model based on neighborhood information.Experimental results show that the proposed method can extract building roofs automatically and precisely,has preferable adaptation to buildings with different complexity,and is able to provide reliable roof information for building reconstruction.4.A building 3D model reconstruction method assisted by building primitive library is improved.Building structure knowledge is integrated into reconstruction through the designed building primitive library.Roof topology graph is constructed by the use of roof intersection line buffer zones,and multiple types of roof constraints are automatic inferred through the matching of roof topology graph and building primitive.Additional constraints least squares fitting algorithm is used to optimize roof parameters globally,ensuring seamlessly connection between roofs in the reconstructed building 3D model.Building feature lines are generated through building boundary point extraction,simplification,and boundary line regularization with multiple principles,and the feature lines combined with the optimized roof parameters are used to create building 3D models.Experimental results show that the proposed method can preferably reconstruct roof topology,and building 3D models reconstructed by the method have seamless connection between roofs.