3D Modeling And Optimization Of Building Based On Remote Sensing Image Scarcity

With the development of remote sensing technology,the 3D modeling and optimization technology of buildings based on airborne optical remote sensing images has important application value in the fields of digital city map update,smart city construction and emergency mapping.The commonly used 3D building modeling methods based on optical remote sensing images at this stage often have high requirements on the number of images and the angles between images,and are not suitable for areas where remote sensing resources are scarce.Although the current 3D modeling method based on a single image can realize the 3D modeling of buildings in the case of lack of images,the single image may have the problem that the building is occluded and it is difficult to observe the complete contour shape of the building,which will affect the modeling accuracy to a certain extent.In addition,it is difficult for the current 3D modeling method to effectively use the outline information and perspective information of the building in the optical remote sensing image to update and optimize the building model,when already having a 3D model of a building with relatively poor quality.In view of the above application background and existing problems,this paper studies the 3D modeling and optimization method of buildings based on airborne optical images.The specific research contents are as follows:1.Aiming at the problem of how to use remote sensing optical images to achieve highquality 3D modeling of buildings when remote sensing resources are scarce,a joint modeling network of two images based on 3D implicit modeling is proposed.In order to better capture the shape and structure features of the building,the attention mechanism is introduced in the encoding stage to enhance the ability of image feature extraction;at the same time,a multiview joint encoding module is added to fuse the features of two images from different angles to enrich the global features extracted by the encoder;in addition,an implicit representation method based on the signed distance field is introduced in the decoding stage,and the signed distance function(SDF)is calculated according to the input features to optimize the surface and the details of the building model and further improve the modeling quality.Compared with the3 D modeling methods in recent years,the network has a better improvement in the accuracy and restoration details of produced building model,and the effectiveness of the improved method has been proved by various experiments.2.Aiming at the problem of how to effectively use the perspective information and outline information of buildings in optical remote sensing images to optimize the 3D model of buildings in the case of existing 3D models of buildings with relatively poor quality,a building model optimization method based on building contour images is proposed.This method extracts building outlines from optical remote sensing images,and then matches the building outline images with the overall outline of the existing 3D model of the building under this perspective.The 3D model of the building is deformed according to the matching difference and constrained by a series of loss function to reduce the field gap and optimize the model details.Finally,the effectiveness of each loss function is proved by ablation experiments,and the comparison experiments prove that the network can be effectively applied to most 3D modeling networks of buildings to obtain better 3D modeling effects of buildings.