AI Generates 3D Lunar Surface Maps... KENTECH Research Team's Paper Accepted by CVPR

Utilizing Images from Danuri and NASA Lunar Orbiters
Up to 10 Times Improvement over Existing Technologies
Core Technology for Lunar Landings and Autonomous Rover Navigation Expected

Researchers at Korea Energy Science University have developed technology that uses artificial intelligence (AI) to generate highly detailed 3D maps of the lunar surface, marking a bold entry into the global space exploration technology race.

The research team led by Professor Lee Seokjoo at Korea Energy Science University announced the development of the AI-based lunar surface 3D map generation technology, the 'LNEM (Lunar Neural Elevation Model),' in collaboration with the Korea Aerospace Research Institute (KARI) and the Korea Astronomy and Space Science Institute (KASI).

This research has been accepted as a regular paper at CVPR 2026, the world’s most prestigious academic conference in the field of computer vision.

3D Lunar Surface Map Generated by AI-Based LNEM Technology and Key Research Areas Restored through LNEM, Along with 3D Terrain Map of the Lunar Surface Created by LNEM.
[Photo by Korea Energy Institute of Technology]

Recently, major space powers such as the United States, China, and the European Union (EU) have been competing in lunar landings and resource exploration, resulting in a growing emphasis on the importance of precise terrain data for analyzing the lunar surface. High-resolution 3D maps are essential not only for selecting safe landing sites but also for enabling autonomous rover navigation and the exploration of space resources.

LNEM, developed by the research team, has been applied to reconstruct the elevation and topography of the lunar surface in three dimensions using actual images taken by NASA’s Lunar Reconnaissance Orbiter (LRO) and Korea’s first lunar orbiter, Danuri (KPLO).

Traditionally, the 'stereo matching' method—comparing multiple images to create 3D terrain—has been predominantly used. However, this approach has limitations in areas with many shadows or lacking clear terrain features, resulting in reduced accuracy.

In contrast, LNEM combines the latest AI technology, neural rendering, with a 'Rigorous Sensor Model' that incorporates imaging conditions and positional information from lunar probes. This enables highly accurate terrain reconstruction even under real lunar exploration conditions.

The research found that LNEM can reconstruct lunar topography at spatial resolutions up to five to ten times higher than existing lunar terrain restoration technologies.

The team also established a data platform called 'Lunar Studio' that integrates images from NASA LRO’s NAC camera and Danuri’s LUTI camera. This platform is designed to allow not only space experts but also AI researchers to easily utilize lunar exploration data.

This research is regarded as a core infrastructure technology for future lunar exploration, going beyond simple terrain reconstruction.

In particular, it is expected to contribute to strengthening the foundation for self-sufficiency in space technology, as it can be applied to Korea’s ongoing lunar exploration projects and international joint exploration initiatives.

Professor Lee Seokjoo stated, "This is a pioneering study that uses images secured by actual lunar probes to restore the lunar surface in detail using AI. We will continue to advance our research so it can be applied to various fields such as autonomous landings, rover navigation, and space resource exploration."

This content was produced with the assistance of AI translation services.