High-precision co-registration of orbiter imagery and digital elevation model constrained by both geometric and photometric information

Abstract

The alignment of images to a reference digital elevation model (DEM) has many applications, especially in planetary exploration. In this paper, we propose a novel high-precision co-registration method for pixel-based matching between an image and a reference DEM to rectify the ever-increasing number of orbiter images to DEMs in a unified reference frame automatically. First, the DEM is converted to a simulated image using a hillshading technique based on a photometric model and the illumination conditions of the image. Then, an initial matching between the simulated and input images is performed based on affine scale-invariant feature transform (ASIFT). The rational function model (RFM) of the image is established and used as a geometric constraint. Then, with tie points (TPs) generated by the initial ASIFT matching, the RFM geometric model of the image is refined, and the gross errors in the TP set are iteratively eliminated. Finally, a high-precision co-registration by pixel-based Least-Squares (LS) image matching is performed using the refined geometric model. Lunar Reconnaissance Orbiter (LRO) Narrow Angle Camera (NAC) and SLDEM2015 (a combined product of Lunar Reconnaissance Orbiter Laser Altimeter (LOLA) and a DEM generated by the Japanese Selenological and Engineering Explorer (SELENE) terrain camera) images are used in our experiment. Results using 17 NAC images at different latitudes demonstrate that the proposed co-registration method is effective and can attain an accuracy of 18.3 pixels in the image space and 25.17 m (0.5 pixel of the reference DEM) in the object space. The method offers an automatic and high-precision matching of LRO NAC images to SLDEM2015. It is also applicable to the co-registration of other orbital images to a reference DEM.