Research On Key Technologies Of Realistic 3D Modeling Based On UAV Tilt Photography

With the continuous development and popularization of UAV technology,UAV tilt photography has become one of the mainstream ways to obtain 3D city models.By using UAV equipped with high-precision cameras and GPS and other equipment,this technology takes continuous aerial photos of target areas and generates 3D models through subsequent processing procedures.It has been widely used in 3D map making,urban planning,cultural relics protection and other fields.There are still some problems in acquisition of image data and the process of constructing 3D models by UAV tilt photography technology.For example,when using a single-lens aerial camera to acquire image data,the fixed angle leads to shooting blind spots;in the production of real-scene 3D models,moving objects(people or cars)in the image will lead to matching difficulties and model geometric deformation and other problems,and weather changes and other factors will make the image have the problem of uneven illumination.In addition,the application of real-scene 3D models is presently mainly limited to the field of appearance display,and has not been fully applied in building monolithic information query and scene roaming.In this thesis,firstly,in the stage of image data acquisition of tilt photography,according to the characteristics of different buildings,an tilt surround route is designed to reduce the shooting blind spots in the processing of image acquisition;secondly,in the stage of data preprocessing,repair and balance the illumination of the image to eliminate the obstructions and uneven illumination,so as to obtain a high-quality and high-precision 3D model of the real scene;finally,the Cesium 3D visualization framework and B/S architecture are used to realize the visualization application of campus spatial data,which can dynamically analyze the 3D model,and provide auxiliary decision support for campus planning.The main research contents of this thesis are as follows:(1)Data collection of UAV.In terms of using UAV to acquire image data,this thesis takes some areas of the Civil Aviation Flight University of China as the modeling area,uses DJI Phantom 4 PRO V2.0 to acquire tilt image data,and designs routes according to the characteristics of different areas,especially for buildings in complex areas,according to the appearance characteristics of objects,design suitable routes for actual shooting,according to construct corresponding circumscribed cuboids and route hemispherical surfaces,and design multi-height,multi-angle tilt surround image acquisition routes,with less images to construct accurate and complete 3D model,thereby improving the efficiency of modeling.(2)Image preprocessing.In order to improve the precision and accuracy of the real-scene3 D model,images are preprocessed before modeling.Since the moving objects in the image will affect the quality of the 3D model,it is necessary to repair the images.An improved GAN network is proposed in this thesis.On the basis of WGAN-GP,U-Net network and CBAM attention mechanism are introduced into the generator;the discriminator with the global discriminator and the local discriminator,and introduces the VGG16 network combined with the SENet channel attention mechanism.The experimental results show that the image restoration effect of the improved network is significantly improved.In addition,the uneven illumination of the image will also affect the subsequent processing effect.An adaptive illumination balance processing method is proposed,which uses fast guided filtering to extract the illumination components,and uses an improved two-dimensional gamma function to dynamically adjust the pixel illumination value to eliminate visual impact of uneven lighting.These preprocessing methods can improve the quality and accuracy of images,and provide a better data basis for real-world 3D modeling.(3)3D modeling and precision analysis.The real-scene 3D modeling software Context Capture is used to automatically process the image data to generate a 3D TIN white model map of the campus,a 3D TIN network construction map,and a 3D modeling result map.Through accuracy analysis,the plane error of the real 3D model is 0.023 meters,and the error in elevation is 0.014 meters,which meets the requirements of actual large-scale measurement accuracy.(4)Visual analysis of 3D models.In terms of model application,the Cesium framework is used to realize the visualization application of campus spatial data through the B/S architecture.Build two functional modules: building individualized information query and scene roaming.Users can query and browse building information on the platform,and also browse the campus scene in the 3D model through the roaming module.The research of this thesis has completed the real-time 3D modeling of some areas of the headquarters of the Civil Aviation Flight University of China.By using the tilt surround route for image acquisition,it is verified that the route has the advantages of high image acquisition efficiency,high model integrity and realistic texture.At the same time,image inpainting method of improved GAN network and adaptive illumination balance processing method are used to preprocess the images,which makes the generated model more accurate and richer in details.Finally,the visualization analysis of the campus is completed,and the building monolithic information query and scene roaming functions are realized.