Map-Guided Coarse-to-Fine Lunar Rover Localization via Multiview Crater Matching

Deep space investigation has gradually developed from remote sensing to planet surface exploration, and rover survey is the main way for planet surface exploration. Precise localization is essential to ensure the safety of the rover. In the past decades, manual-assisted station visual positioning has been extensively employed, while recent studies on track reckoning and visual odometry have provided a foundation for the advancement of automated localization of rovers. However, the cumulative error generated by long-distance exploration is difficult to avoid without global constraints. In this work, we propose a coarse-to-fine localization method wherein craters are employed as landmarks. The orientation of craters captured by the camera serves as positional information to guide a grid search and matching process within the map, achieving initial coarse localization. The precise edges of the craters are extracted from the Panoramic Camera (PCAM) images and aligned with the crater rim in the map to refine the localization result, thereby enabling fine localization of the rover. To validate the proposed method, we conducted experiments on datasets from the Chang'e-3 and Chang'e-4 lunar missions, as well as a simulated dataset containing fifty image pairs constructed from a local digital elevation model (DEM) of the lunar surface. The results demonstrate that, starting from an initial estimate with an error margin of less than 50 meters, our method significantly enhances localization accuracy, achieving meter-level precision, with most results concentrated within approximately 2 meters. Furthermore, the algorithm operates at a sub-second processing speed, which ensures its execution during the rover's panoramic imaging phase between motion and stop, thereby satisfying the real-time localization requirements of actual mission scenarios.