Research On Aerial Triangulation And Dense Matching Of Multi-view Oblique Images

In the era of "Digital Earth",the real three-dimensional model of city,as an important carrier of "Smart City",has shown great potential.Low-level oblique photogrammetry,which can obtain complete ground information from multiple angles,emerges as the times require.It overcomes the defect that the traditional orthophoto can not generate surface texture in the process of true three-dimensional modeling.It has been widely used in large-scale mapping and fast three-dimensional modeling of digital city.Commercial software such as ContextCapture,Pix4 D and Agisoft PhotoScan has been widely used in urban real three-dimensional reconstruction,but there are many problems in image matching,multi-view joint beam adjustment and point cloud dense matching.The traditional orthophoto image data processing algorithm can not meet the requirements of 3D modeling of aerial tilt images.Therefore,based on the existing research results,this paper studies and discusses three aspects of image matching,multi-view aerial triangulation and point cloud dense matching.The main contents of this paper are as follows:(1)Research on oblique image matching,using line-node-line(Line-Junction-Line,LJL)algorithm to replace point matching algorithm,LJL algorithm uses hierarchical strategy to match group and individual segments.The experimental results show that the LJL algorithm is superior to the MSLD algorithm in the oblique image matching with large inclination and wide baseline.(2)Aiming at the problems of large number of images,large overlap and unstable attitude angle in the process of aerial tilt photogrammetry,there is also the exposure delay error caused by the inconsistency of GPS recording time in camera exposure time.Therefore,a joint adjustment model considering the exposure delay error in the multi-view oblique image bundle block adjustment is constructed.In order to ensure the convergence and stability of the bundle adjustment model,the Levenberg-Marquardt optimization algorithm,the sparse matrix algorithm and the weight-selective iterative method are explored to ensure the convergence and stability of the adjustment model.(3)Aiming at the problems of small number of point clouds,low density and many loopholes in the direct application of PMVS algorithm to dense image matching,an improved PMVS algorithm is proposed in this paper.On the basis of least square adjustment,spatial geometric relations and kernel line constraints,the algorithm uses the patch-based multi-photo geometric constraint matching(MPGC)method to optimize the points on the patch.(4)Verify the validity of image matching,combined adjustment model and improved PMVS algorithm in(1)(2)(3),and use the data of Chang'an Campus of Shaanxi normal University to verify.The experimental results show that the algorithm proposed in this paper can be used in the real 3D modeling of aerial oblique images,and the accuracy of the data meets the relevant requirements.