Core Concepts
Developing perception algorithms for lunar polar regions using stereo camera images.
Abstract
I. Introduction
Presents the POLAR Traverse Dataset, high-fidelity stereo pair images of lunar-like terrain under polar lighting conditions.
Aimed at developing and testing software algorithms for lunar exploration.
II. Background
NASA's Moon to Mars program goals include characterizing water deposits on the moon.
VIPER rover mission aims to map water ice distribution on the moon.
III. Setup
Test bed construction with lunar-like terrain and lighting setup.
Use of modified LHS-1 simulant to mimic lunar highlands regolith.
IV. Collection Procedure
Traverses simulated across lunar surface with varying camera parameters and scene conditions.
V. Results & Discussion
3,960 stereo pairs collected across different terrain views and traverses.
Multi-view stereo reconstruction using COLMAP algorithm discussed.
VI. Conclusion
The dataset provides insight into perception algorithms and lighting conditions in lunar polar regions.
Stats
"A total of 3,960 stereo pairs of images were recorded across 4 different terrain views with 6 traverses per view."
"The dataset can be downloaded as a whole (13.4 GB compressed) or as individual terrain view datasets (∼3.3 GB compressed each)."
"Four high resolution LiDAR scans were collected to provide ground truth geometry information."