The study on epipolarity model of satellite stereo-imagery has been the research focus in application field of remote sensing for a long time. So far, there are basically two technical routines for exploring the epipolarity model of satellite images: one totally disregards any geometric sensor model when establishing the epipolar relations but figures out the epipolar curves in image-space by image correspondence and mathematical modeling, e.g. the Polynomial Fitting Model (PFM); on the contrary, the other attempts to exploit the epipolar relations based on specific sensor models, such as the Projection Trajectory based Epipolarity Model (PTEM), the Parallel Projection Transformation based Epipolarity Model (PPEM), etc. Although such models have been studied and used to generate epipolar images of satellite stereo pairs, some technical limitations still exist when taking the universality, applicability and implemental simplicity into account. Accordingly, this paper proposes a novel epipolarity model for satellite stereo-imagery based on the Virtual Horizontal Plane (VHP) of object-space. Firstly, the principle of the VHP-based epipolarity model is described; and then, the workflow of VHP-based epipolar resampling is outlined in detail; finally, to verify the feasibility and correctness of new theory and method, the approximate epipolar images of SPOT5-HRG stereo-imagery are generated. It is demonstrated that the vertical parallaxes of conjugate image points have all reached the sub-pixel level after epipolar resampling; besides, by rearranging the approximate epipolar lines on the VHP defined in our method, the stereoscopic model that is horizontal to the object-space and with consistent resolution can be available.
F. Hu, Mi Wang, Deren Li
Journal name not available for this finding