Determination Of Geometric Model For Geoeye-1Satellite Stereo Images And Analysis Of Accuracy In3D Reconstruction

The technology of the high-resolution remote sensing satellite has been improved rapidly sin ce the21th century, benefiting from the rapid development of computer and material science. At the same time, means of getting large-scale topographic maps is no longer restricted to th e aerial photogrammetry; the aerospace photogrammetry has been widely used. Further, the t echnology of the high-resolution remote sensing satellite has superior capacity in data collect ion and transmission, and is free of air traffic control. With these merits, this technology is p1aying an an ever increasing important role in various areas of the national economy.However, rigorous sensor models for the remote sensing satellite are difficult to be obtained for several reasons. Firstly, the high-resolution remote sensing satellites has a long focal leng th and a narrow viewing angle since they travel in a high sun-synchronous orbit and still mai ntains high resolution. Secondly, these satellites usually use the linear push-broom CCD sens or, whose imaging mechanism is very complex, and there exists high correlations among sen sor orientation elements. Finally, commercial satellite companies generally do not provide us ers with parameters of rigorous image sensor models for confidential considerations.As an alternative to the rigorous sensor model, the rational function model (Rational Functio n Model, RFM) is widely used in high-resolution remote sensing satellite image positioning t o simulate the mapping from the image coordinates to the object coordinates.This thesis discussed solution methods of the coefficients in the rational function model cons idering the locating features of the high-resolution satellite image sensor model. Solution me thods of two types of rational function models, i.e., the terrain related model and the terrain u nrelated one were derived. Further, two optimized error compensation methods were propose d based on the error features of the two rational function models. Then, the validity and accu racy of both error compensation methods were validated through experiments.In the terrain unrelated model, coefficients of the rational function were calculated through fi tting the image spatial grids, which in turn were obtained through the rigorous sensor model. Although this approach is very accurate, systematic error frequently appears due to the absen ce of corrections from ground control information. To overcome this shortcoming, an optimi zed compensation method in image space was proposed in this thesis. Experiments have sho wn that this improved method has high accuracy both in locating and3D reconstruction.The terrain related model uses ground-pixel control points as the solver benchmarks. When t he ground is relatively flat, coefficients of the rational function model can be obtained in hig h accuracy. However, in uneven areas, the density and distribution of the ground control poin ts are affected by the undulating terrain; the errors in solving the model coefficients in areas f ar away from control points may increase. This kind of error is highly random and difficult t o be corrected.This thesis investigated patterns of the rational function model from the point views of their mathematical properties using the "simply-recovery" technique. Based on the simulated dat a, a compensation method to the rational function model was determined. Experiments of the terrain related approach shown the high accuracy of this compensation method.In addition, a3D reconstruction method based on the rational function model was derived thr ough the intersection process of the rational function model. The accuracy of this reconstruct ion method was tested using a GeoEye-1Satellite Imagery. Also, a3D mapping was carried out using this method. Finally, the accuracy of the topographic mapping was assessed based on the external precision DEM.