| With the rapid development of smart city,virtual reality,laser scanning and other technologies,major cities have rushed to bring digitization and information into the construction of the city.The virtual reconstruction of large-scale urban scenes plays an important role in urban planning,smart transportation,military simulation drills,and urban post-disaster reconstruction due to natural disasters or human-attacks.With the improvement of technology and city information,the three-dimensional modeling of large-scale urban scenes will play an increasingly important role.The urban scene is filled with buildings.The three-dimensional reconstruc-tion of a building has always been a hot research field in computer vision,com-puter graphics,and surveying and mapping.However,there have been many difficulties,such as the accuracy problem of the current three-dimensional scan-ner leads to the general noise of the acquired data,many automated modeling algorithms cannot solve the complex building structure.This paper addresses the current challenges in the field of building reconstruction and the current strong demand for urbanization in this area.This paper first,adopts an approach tak-ing into account both the distance and normal variations to solve the artifacts of point cloud data commonly acquired by laser scanners.A region growing method is used as the point cloud segmentation algorithm.Experiments show that the segmentation algorithm can effectively resist a certain degree of noisy data,and the segmentation algorithm in this paper has a better segmentation effect than the traditional RANSAC-based segmentation algorithm.Then,aiming at the common characteristics in urban buildings,this paper first fits an initial build-ing model through the segmentation obtained in the point cloud segmentation stage,and then optimizes the model from three aspects of orientation,position,and isometric relationship.After each stage of optimization,a new parameter set of the model is obtained by solving a nonlinear optimization problem with constraints.In this paper,the distance error between the building model and the segmentation point cloud data is used as the objective function of the opti-mization problem,which largely improves the matching of the facade data and point cloud data of the building.A large number of experimental results show that a better building model can be obtained by optimizing the initial building model through the three stages of this paper,which demonstrates the effectiveness of the proposed algorithm. |
