Automatic Georeferencing of Aerial Images Using High- Resolution Stereo Satellite Images

For airborne surveys, direct georeferencing has become the primary source for EOPs (Exterior Orientation Parameters) determination since integrated GPS/INS (Global Positioning System/Inertial Navigation System) systems were introduced. However, there is still need for alternative indirect georeferencing since there are remote, inaccessible areas that lack a geodetic infrastructure and thus GPS/INS-based georeferencing is not feasible. In addition, terrain-referenced navigation is gaining momentum, where the assumption is that no GPS is available. High-resolution satellite images have been globally available and newer high-resolution satellite images offer not only better spatial resolution with decreasing revisit time, but high positional accuracy up to subpixel if ground control is available. Therefore, high-resolution satellite imagery has high potential as a ground control source for aerial image georeferencing and terrain-referenced navigation. Indirect georeferencing of aerial images usually requires accurate 3D ground control points. Unfortunately, ortho-rectified imagery, which is conventionally used as a reference for image-to-image georegistration, contains relief displacement due objects on the ground, resulting in horizontal errors; note that accurate DSM (Digital Surface Model), including terrain and all features on the ground, is usually not available globally. In this study, a high-resolution stereo satellite image-based automatic georeferencing approach is proposed. The use of stereo images can avoid the impact of relief displacement and requires no DSM to obtain ground heights. The matching between aerial and satellite stereo images is based on the SIFT (Scale-Invariant Feature Transform) features, and outliers are pruned utilizing RANSAC (RANdom SAmple Consensus). Finally, to recover 3D ground coordinates, cross correlation matching is performed for epipolar resampled satellite images using FFT (Fast Fourier Transform). An experiment was carried out for a strip of aerial images, including various terrain textures, and showed good potential for the approach.