Research On Stereo Matching Technology Of High Resolution Generalized Satellite Images

With the rapid development of remote sensing and space technology,high resolution satellite image has become an important tool for acquiring information on the surface of the Earth.With the increase in the number and resolution of satellites,it is possible to use the high resolution satellite imagery to acquire the Digital Elevation Model(DEM)or to reconstruct the ground target in three dimensions.In this paper,generalized satellite image pairs(Gaofen-1,Gaofen-2,Pleiades-1,Pleiades-2,QuickBird,WorldView-2,WorldView-3)are used for feature extraction and matching,and then the rational function front intersection model is used to obtain the ground point 3D,as well as to coordinate and generate a digital elevation model,and finally the accuracy of the results obtained.The experimental results show that the method is reliable and effective.The main contributions of the paper are as follows:The rational function model is analyzed,and its basic formula and method for solving the model parameters are given.In addition,the coefficient fitting error of the rational function model of GF-1,GF-2,Pleiades,QuickBird,WorldView-2 and WorldView-3 images was analyzed.The feature extraction of generalized satellite image pairs is performed using the KAZE algorithm and improved for the KAZE algorithm,The improvement idea is to preserve the advantage that KAZE algorithm can extract stable feature points,and optimize the running time of the algorithm: add second derivative in KAZE descriptor to enrich image information.The circular area is used to replace the rectangular area of the original algorithm,and a 40-dimensional vector is constructed by redividing 5 areas,and it is not limited to the Euclidean distance.Then,the time efficiency of the improved algorithm and KAZE algorithm is analyzed by strict mathematical relations,and the improvement is obtained.The algorithm is superior to the original algorithm in most cases.Under the conditions of Gaussian blur,noise,rotation scaling,illumination,etc,the stability of SIFT,SURF,KAZE and improved algorithm are analyzed.The conclusion of this paper is that KAZE and its improved algorithm are better than SIFT and SURF algorithms.The grid-based motion statistics(GMS)algorithm is used to perform feature point matching.The experiment also shows that the matching rate of the algorithm is very high.Then the obtained point of the same name is substituted into the intersection model of the rational function,and the three-dimensional coordinates of the ground point are obtained.The error compensation is performed on the rational function model,and the appropriate three-dimensional coordinates are obtained.Finally,the accuracy of the method is obtained by comparing with the coordinates of the real control point.The accuracy of the GF-1 generalized image pair is 83%,the accuracy of the GF-2 generalized image pair is 78%,the Pleiades generalized image pair accuracy is 76%,the WorldView-3 stereo pair precision is 85%,and the drawing is done.Three-dimensional scatter plots and corresponding error plots.