Geolocation bias correction for CARTOSAT-1 stereo images through virtual ground control points generation over mountainous areas

Rational Polynomial Coefficients (RPCs) are commonly used for georeferencing satellite images. However, sometimes RPCs cannot accurately establish a ground-to-image relationship, resulting in what is known as RPC bias. This bias is corrected using existing 3D data such as Digital Elevation Models (DEMs), point clouds, topographic maps, or 2D data such as maps, georeferenced images, or orthophotos. This paper focuses on using CARTOSAT-1 satellite stereo images and proposes a DEM matching-based method for correcting RPC bias in these images. Firstly, it is demonstrated that RPC bias should be corrected in image space. Secondly, a satellite-based DSM is generated from the CARTOSAT-1 stereo images and then aligned with an ALOS PALSAR DEM using mutual information. The extracted correspondences through mutual information are then back-projected into image space. By fitting a 2D affine model to these correspondences, the RPC bias is corrected. Moreover, the accurate ground-to-image correspondences are considered as virtual Ground Control Points (GCPs). The proposed method is evaluated on three different datasets using some distributed GCPs. After correcting RPC bias, the Root Mean Square Error (RMSE) of the proposed method is found to be less than 5.9 m for georeferencing CARTOSAT-1 images. The results also demonstrate the superiority of the proposed method compared to other RPC bias correction methods.