The Method Of 3D Fine Modeling Of Buildings Based On Unmanned Aerial Vehicle Oblique Photography

For large-scale urban scenes,building a three-dimensional city model quickly,accurately,and efficiently is one of the key points in the construction of a "smart city".Buildings are the most important features in urban scenes,and are the main places for human activities.The 3D building model is an integration of building geometry,texture,and attribute information.It has great potential in many applications such as urban spatial information management and urban planning and construction.Unmanned aerial vehicle(UAV)oblique photography has the advantages of simple data collection,low modeling cost,and fast modeling speed.It is an important method for urban scene data acquisition and high-precision 3D model construction.It can quickly generate a 3D geographic information model of the overall scene.The constructed 3D model has strong sense of reality and rich texture,which is an ideal way for 3D modeling of large-scene cities.However,the three-dimensional model generated by the automatic processing of the oblique photography image of the drone through the oblique photography software is a geometric model of the entire scene,which cannot distinguish different features,and cannot analyze and manage buildings and other features.The current UAV oblique photography building single modeling method has a large workload and low efficiency,and it is difficult to obtain a fine 3D model of a building with a high level of detail,and cannot be applied to the construction of a 3D model of a building in a large-scale urban scene.Based on the above problems,this paper uses the digital orthophoto of the scene and the multi-view dense matching point cloud acquired by oblique photography software as the experimental data.The goal is to use the image segmentation convolutional neural network to quickly and accurately extract the building objects from the orthophoto of the scene;Optimize the building outline based on the extracted building objects;accurately extract the pure building single point cloud from the overall point cloud of the scene based on the optimized building outline;construct the threedimensional fine building model based on the building single point cloud.Based on the objectives of this article,relevant research has been carried out.The main research contents are as follows:(1)Digital Orthophoto Building Extraction Based on Convolutional Neural NetworkThe image segmentation network in the convolutional neural network is introduced into the oblique photographic image building extraction,the convolutional neural network hierarchy and the image segmentation network model are analyzed and researched,and the building extraction data set is made by the data annotation tool for the digital orthophoto of the scene.Based on the data set constructed in this paper,the four image segmentation networks(UNet?Seg Net?ENet?ERFNet)are compared in the same experimental environment,and the network model with the best extraction effect of oblique photographic image buildings is selected.Based on the best model parameters of the best network obtained by training,the automatic extraction of digital orthophoto buildings in the target area is completed.(2)Feature optimization of building contourThe building object acquired based on the image segmentation network is composed of arbitrary polylines,and the degree of right angle is low,which does not conform to the real building outline.This paper uses the closed operation in the mathematical morphology method to optimize the concave part of the contour and smooth the boundary of the building;extract the contour line of the building object through the Canny operator;Delete,use Douglas–Peucker polygon approximation algorithm to delete redundant points of contour.This paper proposes a building outline optimization method based on the main direction of the building and the center of gravity of the building outline to delete the redundant points of the outline and the right angle processing of the characteristic edges,and generate the outline of the building more in line with the real scene and geographic cognition.(3)Refined modeling based on single building point cloudUsing multi-level details(Level of Details,LOD for short)technology can express different levels of fineness of the geometric model of the building.Based on the obtained optimized building outline,a pure single building point cloud is extracted from the overall scene point cloud.Based on RANSAC point cloud segmentation,the building point cloud is divided into main structures such as roofs and facades;based on the main structure,detailed structures such as building eaves,parapets,and balconies are extracted.Based on the elevation position parameters obtained by RANSAC point cloud segmentation,the normalized facades point cloud of the building is re-obtained.Based on Delaunay triangulation,the main structure of the building and the detailed structure are generated to generate a triangulation geometric model,and the geometric models of different parts are spliced to obtain a LOD2 level fine three-dimensional model of the building.Through the processing of the method in this article,based on the digital orthophoto of the scene and the multi-view dense matching point cloud data generated by the oblique photography software,a refined three-dimensional model of the building can be constructed to provide support for the refined management of the "smart city".