New Techniques for Image-based Planetary Topography and Shape Modeling

Kurzfassung

In recent years, new deep learning techniques have been applied in various areas of lunar and planetary science. Here, we introduce their application to planetary topography and shape modeling based on optical observations with the focus on the Moon. To efficiently handle the rapidly growing volume of meter-scale high-resolution imagery, we employ monocular depth estimation, aided by low-resolution topography, to recover high-quality terrain details. In the presentation, we will demonstrate the capabilities of the proposed approach under varying illumination conditions for different regions of interest characterized by a variety of landforms. We will also evaluate the processing efficiency and compare it with traditional image-based topographic modeling techniques, including stereo-photogrammetry and shape-from-shading. The evaluations are conducted using Lunar Reconnaissance Orbiter Narrow Angle Camera (LRO NAC) images and Lunar Orbiter Laser Altimeter (LOLA)-derived topography. Furthermore, the different application-oriented needs of large planetary bodies (e.g., the Moon) vs. small bodies (e.g., Itokawa and Ryugu) are discussed. We will also briefly introduce Neural Radiance Fields (NeRF) adapted for high-fidelity small-body shape modeling using only sparse images sets.