| As the data base for the construction of 3D real China,smart city and digital twin city,3D building model has been widely used in the fields of urban management and planning,real estate management and ancient architectural culture protection.At present,laser LiDAR point cloud,as an efficient and accurate means to obtain 3D information of ground objects,has become an important data source for 3D reconstruction of buildings.In the follow-up processing of point cloud data,due to the limitations of the lack of semantic information,large redundancy and irregular distribution of the data itself,the relevant algorithms of point cloud data processing often have problems such as low efficiency and difficult to ensure accuracy.In addition,due to the limitation of field Angle and occlusions of ground objects,the reconstructed3 D model also has some problems,such as incomplete geometric structure and missing texture.Therefore,this study focused on LiDAR point cloud data processing and modeling technology,and studied and discussed key issues such as point cloud filtering,point cloud registration and multi-source data fusion modeling.The main research work and conclusions of this paper are summarized as follows:(1)This paper analyzes the current research status and shortcomings of three-dimensional building modeling;This paper introduces the related concepts and principles of 3D laser scanning technology,point cloud filtering and point cloud registration algorithm,including the working principle of 3D laser scanning technology,the characteristics and preprocessing of point cloud data,the principles and steps of classical point cloud filtering and registration algorithm.The theoretical knowledge of building 3D reconstruction is summarized.The feasibility of multi-source data fusion modeling is explored.(2)Aiming at the problems of large degree of manual intervention,single application scenario and poor adaptability of threshold in traditional point cloud filtering algorithm,this paper constructs an improved adaptive threshold filtering algorithm.The algorithm is based on quadric surface fitting filtering algorithm and adopts multistage filtering scheme.Firstly,the virtual grid is introduced to segment the preprocessed point cloud data,and the seed points are selected according to the neighborhood grid.Secondly,the surface fitting parameters were calculated by mixed least square method.Then,the difference between the real elevation and the fitting elevation is calculated,and the filtering threshold is determined by combining kmeans clustering and normal distribution.Finally,the multistage filtering strategy is used to change the grid size step by step to get good filtering results.Based on the public data released by ISPRS and the airborne point cloud data collected by ISPRS,the experimental research is carried out and compared with the classical algorithm provided by ISPRS,the conventional mobile grid surface fitting filtering algorithm and the CSF algorithm.The results show that the proposed algorithm can retain the surface detail features well,and the algorithm is more stable.For data set(1),the mean total error of the algorithm in this paper is 7.04.For data set(2),the total error of the algorithm in this paper is 4.02.The filtering result of data set(1)is better than the seven algorithms of comparison,and the filtering result of data set(2)is better than the two algorithms of comparison.(3)Aiming at the problems of the traditional iterative nearest point(ICP)algorithm,such as low efficiency,large amount of calculation,long convergence time and easy to be affected by the initial pose,a 3DHarris+SAC-IA+ICP point cloud registration algorithm was constructed in this paper.The algorithm adopts the registration strategy of "combining thickness and fine".Before registration,3D-Harris feature point detection algorithm is used to extract the feature points of source point cloud and target point cloud,in order to preserve the original features of point cloud and minimize the amount of registration calculation.Secondly,the fast point feature histogram(FPFH)features of the feature points are calculated.Then,based on this feature,the sampling consistency initial registration(SAC-IA)algorithm was used to complete the initial registration.Finally,ICP algorithm is used to complete the registration on the basis of rough registration of good posture.Based on the public point cloud data released by Stanford University and the collected point cloud data,this paper conducts a controlled experiment.From the two indexes of time consuming and mean square error,the proposed algorithm,the classic SAC-IA+ICP algorithm and SAC-IA combined with NDT are compared and analyzed experimentally.The results show that: In this paper,the algorithm has better performance in the calculation efficiency,registration accuracy and processing effect.The average registration time performance of the proposed algorithm is improved by 44.56% and 32.89%,respectively.The average registration error performance increased by 52.57% and 30.68%.(4)Aiming at the problems of incomplete model and low accuracy in the modeling of a single data,this paper carried out a research on three-dimensional building modeling based on the point cloud data generated by tilt photogrammetry and three-dimensional laser scanning point cloud data,and on the basis of the fusion of the two kinds of point cloud data.When using the fusion data for 3D modeling,firstly,the UAV tilt image data was processed by Contextcapture software to generate intensive point cloud data.Then,the registration algorithm constructed in this paper is used to fuse the dense point cloud data with the measured laser point cloud data after processing.Finally,the three-dimensional building modeling is carried out,and the results are compared with the three-dimensional building modeling results using a single data.The results show that: compared with the modeling results using a single data,the threedimensional building model generated by multi-source fusion data is more accurate,complete and beautiful. |
PDF Full Text Request