Research On Key Technology Of Point Cloud Generation With Remote Sensing Imagery Based On 3D Reconstruction

With the development of remote sensing technology and computer vision technology,people’s demand for geographic information in some fields has gradually extended from two-dimensional(2D)image data to three-dimensional spatial information,and point cloud has become the main data format to represent the three-dimensional(3D)world.Multi-view remote sensing images with high resolution and wide coverage can be used to obtain high-precision and high-integrity terrain point cloud data through intensive matching and point cloud optimization and other processing technologies,which can effectively realize the expression of 3D spatial structure information and 3D reconstruction in a large range of terrain areas.In this thesis,the remote sensing technology and the computer vision technology are used to obtain and optimize the three-dimensional point cloud of large-scale terrain area of remote sensing image.The main contents of the research are as follows:(1)Remote Point cloud generation of remote sensing image based on improved Patch-Based Multi-View Stereo(PMVS)algorithm.The PMVS algorithm is a multi-view stereo algorithm based on patch,which can efficiently reconstruct a 3D point cloud model of the scene from a set of 2D images with known camera parameters.In this thesis,an improved multi-view stereo algorithm based on patch is proposed for 3D reconstruction of multi view remote sensing terrain image with large width and obvious weak texture region.The algorithm firstly uses the concurrent Scale-invariant feature transform(SIFT)operator based on image blocks to extract the feature points of the image,and then calculate the initial seed patches through the matching propagation strategy based on the ground elevation range constraint,and finally obtain the three-dimensional dense point cloud of the image corresponding to the ground object scene through the iterative process of the expansion and filtering of patches.The experimental result shows that the improved PMVS algorithm in this chapter can reconstruct the dense point cloud with higher integrity by using the multi-view remote sensing terrain images with large width and weak texture,and the time efficiency of reconstruction is also guaranteed.(2)Point cloud optimization of image based on improved the Iterative Closest Point(ICP)registration algorithm.And in order to further optimize the integrity of the point cloud which is generated by improved PMVS algorithm,this thesis uses point cloud of multi-group images generated by multi-view remote sensing images in the same terrain area for registration and fusion.However,the difference of initial pose between the point clouds of image is large,and the point clouds are disorderly distributed in space,and the number is large,so it is necessary to carry out high-precision registration work.According to the characteristics of point cloud of image,this thesis proposes an image point cloud optimization method based on the Grey Wolf Optimizer(GWO)and improved ICP algorithm.The whole algorithm is divided into two stages: the initial registration stage of point cloud of image based on GWO algorithm and the fusion stage of point cloud of image based on improved ICP algorithm.The experimental result shows that the method proposed can accurately fuse multiple sets of point cloud on the basis of accurate registration,and can effectively optimize the point cloud.(3)Design and implementation of the point cloud generation system of remote sensing image.The system of point cloud generation of remote sensing image is designed and implemented based on the improved multi-view algorithm and the point cloud and optimization algorithms.The system is re-developed on the open source project Map Win GIS,and the related algorithms are packaged into dynamic link library files and implemented in the main window program of the system in the form of plug-ins to realize the transformation from the 2D information of remote sensing images to the 3D point cloud information,which can provide high-precision 3D point cloud data for related applications.