Projective Invariant Triad Indexing for Integrated Lunar Terrain Relative Navigation

Hyunsung Kim, Sunwoo Mun, Seokju Lee

Keywords: projective invariant indexing, terrain relative navigation, lunar landing mission

Abstract: Robust lunar terrain relative navigation (TRN) requires bridging the severe modality gap between offline heterogeneous orbital payloads and runtime monocular descent imagery. To seamlessly integrate these disparate views, we construct a 3D geometry-aware projective invariant triad index. By lifting 2D geo-referenced crater rim ellipse parameters into 3D disk quadric primitives, we extract appearance-independent geometric invariants that are independent of illumination and tolerant to viewpoint changes at the representation level. To evaluate how these properties propagate to trajectory-level performance, we tightly couple this index with a visual odometry (VO) module via an extended Kalman filter (EKF). Experiments using a QuickMap-based lunar south pole simulation systematically map the representation's operational envelope across observation noise, illumination, and off-nadir angles. Results show that within this envelope, sparse absolute geometric fixes suppress accumulated VO drift in mean position error by over 88%, demonstrating an integrated framework for continuous downstream lunar descent navigation.

Submission Number: 9