To ensure long-term space missions, an autonomous visual navigation
system for lunar rovers supporting selfexploration is demanded. While
the object-centric localization and mapping problem can be solved
through state-of-the-art methods, they greatly rely on human-in-loop
remote operations, posing several challenges for the visual system of a
rover when operating in a distant, unknown, and feature-sparse lunar
environment. This paper presents a SAM-augmented object-centric SLAM
framework which enable rovers to estimate the relative distance to the
target on lunar surface, thus ensuring the safety of exploration task.
Based on the feature-matching baseline and an auto-label segmentation
approach, a prompted-based object instance extraction pipeline is first
proposed to predict object correspondences at a pixel level. We then
integrate with front-end outputs in the middle-end of SLAM to tightly
associate reliable object-centric constraints between image frames.
Moreover, the data association in LO-SLAM maintains robust camera-object
relative position estimation between the camera and target object.
Extensive experiments are conducted on our dataset, Stereo Planetary
Tracks (SePT). Results show that the proposed LO-SLAM is validated on
challenging lunar scenarios with dramatic viewpoints and object scale
changes. The average pose errors are less than 0.37 meters in centroid,
and the average object-centric trajectory error is less than 0.49%. An
open-source dataset has been released at
https://github.com/miaTian99/SePT_Stereo-Planetary-Tracks.