VPR-Space benchmark for visual place recognition in space analog domains
VPR-Space evaluates visual place recognition capabilities on extraterrestrial surfaces.
VPR-Space is a novel benchmark to evaluate Visual Place Recognition (VPR) capabilities on extraterrestrial surfaces. This benchmark uses sixteen VPR database-query datasets. These datasets come from Space Analog Domains (SAD).
SAD are terrestrial environments that resemble properties of the Moon or Mars. VPR is an important means for loop closure detection in visual Simultaneous Localization and Mapping (SLAM).
The researchers propose a novel approach to generate ground-truth information. This decision depends on whether a match would be valuable in the downstream SLAM task. They use ground-truth (GT) matrices based on an ensemble of state-of-the-art Relative Pose Estimators (RPE).
The datasets include challenges regarding environment, viewpoint change, and illumination conditions. Existing VPR and SLAM approaches excel in urban and feature-rich natural environments. However, extraterrestrial surfaces like the Moon or Mars pose severe challenges.
Nine state-of-the-art VPR approaches were evaluated. The results show considerable room for improvement. These results indicate directions for future research.
Key facts
- VPR-Space uses sixteen VPR database-query datasets from Space Analog Domains (SAD).
- The benchmark evaluates VPR capabilities on extraterrestrial surfaces like the Moon or Mars.
- Ground-truth (GT) matrices are based on the performance of an ensemble of state-of-the-art Relative Pose Estimators (RPE).