Research On Stereo Matching And Reconstruction For Building Areas Of Multiangle Remote Sensing Imagery

With the maturity of satellite stereo imaging capability,the application requirement of stereoscopic satellite data has attracted more and more attention,meanwhile the satellite data has been upgraded from traditional stereo imagery to multiangle remote sensing imagery.Different from traditional two-dimensional graphic imagery,the stereo imagery and corresponding parameter files can be used to achieve 3D reconstruction for object areas.As one of the most common objects in human life,buildings plays an important role in our lives.Therefore,the thesis in our paper is mainly focused on sensor imaging model,joint stereo matching and reconstruction of multiangle remote sensing imagery.the rational function model is mainly discussed among the generalized satellite imaging model,and some technique means are used in stereo matching process according to the own characteristics of buildings.Based on these,the least square method of the rational function model and DSM fusion and interpolation technique are introduced to achieve the three-dimensional reconstruction of building area.In the whole stereo reconstruction system,the imaging model of remote sensing imagery establishes the coordinate relation of the 2D image coordinate to the 3D geodetic coordinate.Started with the most common generalized sensor model—the rational function model,we analyzed its advantages and disadvantages,and introduced two classical model optimization algorithms.Furthermore,considering in the insufficiency and non-uniform distribution of ground control points in the process of optimization process,some methods are proposed.Finally,combined with classical algorithm and the virtual ground control points,a method is proposed to update the rational polynomial coefficients which can be used during epipolar imagery generation and three-dimensional reconstruction process.Before studying the stereo matching algorithm,in order to shrink the searching space,epipolar constraints is added to the test data to guarantee all the homonymous points located in the same line.After that,we set the middle imagery as based imagery to search matching points in object imagery.Taking consideration of abundance corner points and edge lines in building areas,point feature extraction and matching technology are adopted to determine the object imagery matching search range,then an improved line extraction algorithm is proposed.Due to the line disparity and location distribution,we gained the maximal and minimum disparity searching range of pixel among all the image space.Then area-based matching process is carried on restrained by line constraint disparity and dense disparity maps are gained.Finally,in order to achieve stereo reconstruction for building areas in the real scene,the least squares algorithm based on rational function model is analyzed so that we can transform the matching disparity map into 3D information and obtain dense DSM point cloud data.Then,due to the incompleteness during matching process,DSM fusion and interpolation technology are also utilized.After those,experiments are verified compared with the Lidar data of the same test area and have a good result.