3D registration of remote sensing data for planetary exploration

Loïs Brun, Sylvain Doute, Jeronimo Bernard-Salas, Adeline Paiement

Keywords: 3D rigid registration, remote sensing, terrain modeling

TL;DR: A 3D geometry-based method for aligning remote sensing observations that outperforms existing approaches in case of partial overlap and heterogeneous resolutions.

Abstract: Data registration is a crucial preliminary step in exploiting complementary remote sensing data from different satellites. In planetary science, registration is notoriously difficult due to a lack of salient landmark features and strong discrepancies in the spatial extent (denoted as footprint) of acquisitions. We propose to address this issue using 3D information derived from pairs of images, available as Digital Terrain Models (DTMs). We propose a method based on 3D geometric information for the rigid registration of DTMs. Our method is comparatively evaluated on a new benchmark dataset of synthetic planetary DTMs, as well as on real Martian data, against baseline and state-of-the-art methods. It demonstrates a satisfactory accuracy, while proving significantly more robust to common remote sensing footprint challenges such as missing data and varying amounts of overlap.

Submission Number: 5