Research On Building Point Cloud Fusion And 3D Reconstruction Method Based On Ascending And Descending Orbit TomoSAR

Synthetic Aperture Radar(SAR)tomography is a three-dimensional imaging technique that uses repeated return SAR images of multiple views of the same target to achieve threedimensional point cloud imaging of the observation area.Building reconstruction using highresolution spaceborne TomoSAR is of great significance and potential in urban planning and urban modeling.Since the imaging mode of SAR is side-by-side imaging,the TomoSAR point cloud of a single orbit cannot achieve complete observation of buildings,while the TomoSAR point cloud comes with unavoidable noise point clouds in the generation process,making it difficult to directly apply the TomoSAR point cloud to the production practice of live 3D in urban areas.However,the existing methods are difficult to extract the same features as well as to use the overlap rate to achieve the alignment of different orbital TomoSAR point clouds,as well as the existing TomoSAR point cloud processing techniques rely on data segmentation,which greatly reduces the efficiency and accuracy of processing.In this paper,we investigate the technical barriers to the application of TomoSAR point clouds in urban buildings.In terms of TomoSAR point cloud fusion,we study a coarse and fine alignment strategy for TomoSAR point cloud fusion in urban buildings;in terms of TomoSAR point cloud 3D reconstruction,we study a neural network-based TomoSAR point cloud regularization method.The research and work in this paper are divided into the following two main sections:(1)For the problem of difficult feature extraction and very low overlap rate between the ascending and descending TomoSAR point clouds,a coarse and fine alignment method is proposed in this paper.First,statistical filtering is used to eliminate most of the noise points and outliers,and then the Do PP projection is used to extract the TomoSAR building facade point cloud,and the facade points for subsequent alignment parameter calculation are obtained based on density clustering.Then,the rough registration of the two point clouds is carried out based on PCA.Finally,the rotation and translation correction are carried out by using the angle of the normal vector of the parallel facade of the building and the distance of the outer end of the facade projection.The experimental results verify the feasibility and robustness of the TomoSAR point cloud fusion method in this paper.for the fusion of ascending and descending TomoSAR point clouds,the experimental results show that the average rotation error is less than 0.1 °and the average translation error is less than 0.25 m.For cross-source TomoSAR point cloud fusion,the registration accuracy evaluation results are the defined angle and distance,which are less than 0.2 °and 0.25 m respectively.(2)Aiming at the irregular surface of TomoSAR building point clouds and the low automation of its point cloud processing method,this paper investigates a fully connected neural network-based method for three-dimensional reconstruction of TomoSAR point clouds.Inspired by the regression analysis,we project the point cloud along the direction of the satellite incidence angle in order to apply it to the neural network’s prediction of the point height.The height of the points is changed by the designed neural network,which allows the points on the building surface to be refined.The method accurately preserves the original structure of the building while regularizing the building surface.Thanks to the regression mechanism,the method is highly automated and avoids data segmentation and complex parameter adjustment.The experimental results show that the method has high accuracy and robustness in denoising and regularizing TomoSAR point clouds of urban buildings.