Research On The Improvement Of Georeferencing Accuracy Of High Resolution Satellite Imagery

With the development of satellite technology,computer science,and sensor technologies,high resolution satellite imagery(HRSI)has been considered as a more important role in the national economic development.Georeferencing technique has been a key step for the basic geographic data generation by using HRSI.The principle and methods of georeferencing are researched in this dissertation for accuracy improvement,in which the error-in-variables model and multisource geographic data are introduced for georeferencing as key techniques.The aim of our work is to provide a reference for global geographic data generation.The main works and innovations can be summarized as follows:1.The error-in-variables(EIV)model is introduced for adjustment computation.The solutions to total least squares,the virtual observation based total least squares and the regularized total least squares are deduced based on the EIV model respectively.And these algorithms are validated by experiments.2.Since both the design matrix and observations contain errors in the Rational Function Model(RFM)bias compensation procedure,a total least squares based RFM bias compensation method is proposed in this dissertation,which could improve the adjustment computation accuracy and the georeferencing accuracy without introducing extra control information.3.For the reason that the design matrix of image point observation contains errors while that of virtual observation matrix is error-free,the virtual observation based total least squares methods are introduced for bundle adjustment problems based on rigorous sensor model and rational function model respectively.The experimental results indicate that the proposed methods could improve the accuracy of bundle adjustment.4.The georeferencing methods for HRSI by straight control lines are investigated in this dissertation.The EIV model is introduced and a total least squares method is proposed for bundle adjustment based on line features.Conjugate point coincidence constraint is introduced to improve the georeferencing accuracy when the quantity of straight control lines is not enough.The experiments illustrate that the line feature based method could achieve an equivalent result when compared to the method based on point feature.The proposed method could improve the line based georeferencing accuracy and provide a new way for georeferencing processing in the areas that are rich in line features such as urban areas.5.The global Digital Elevation Model(DEM)datum is introduced for georeferencing accuracy improvement.A surface matching algorithm is deduced to minimize the Euclidean distance between the relative surface and the reference DEM.An approximate solution to the normal vector of relative terrain is deduced based on the local similarity,after which a sparse surface matching algorithm aiming to estimate the transformation parameters between tie points and reference DEM is proposed in this dissertation.The proposed method could reduce the quantity of tie points for parameters estimation and be used in areas where the terrain fluctuates gently(except for areas such as steep and high mountains).Since traditional surface matching algorithm is based on the similarity transformation that would result in ill-posed problems,the 6-paramter model,the 5-parameter model and the 3-parameter model are analyzed.It is concluded that the 3-parameter model is more suitable for terrain matching when the georeferencing accuracy of HRSI is good and the rotation error in object space is little enough.Considering that the design matrix of three parameters model contains errors,the EIV model is introduced for parameters estimation.After surface matching,a method for ground control points extraction from DEM is proposed in this dissertation.The experimental results indicate that the proposed method could improve the georeferencing accuracy without field measured ground control points,and would provide good initial values for georeferencing with several ground control points.6.The Google Earth datum is introduced for georeferencing by HRSI,and the results by different controlling strategies are analyzed.The accuracy of ground control points extracted from Google Earth in three areas is also checked.The experimental results indicate that the quantity of ground control points(GCPs)could be reduced with Google Earth in georeferencing processing,and our proposed methods would work when the quantity of GCPs is not enough or the distribution of GCPs is not good.