LUSE-MSCKF: A Landmark-Utilizing Enhancement of MSCKF for Robust and Efficient Monocular Visual-Inertial System

When GPS signals are unreliable, robots must achieve autonomous localization using onboard sensors. Due to the limited computational resources of unmanned systems and the need to allocate these resources across multiple tasks, it is essential to enhance localization efficiency while maintaining accuracy. This paper proposes a landmark optimization method based on null space projection. A residual model is constructed from visual-inertial measurements, and equivalent residual equations for the state and landmark information are respectively constructed via null space projection. Extended Kalman Filter (EKF) updates are then performed separately on the state and landmarks. Experimental results on the public dataset (TUM-VI) demonstrate that the proposed method achieves superior localization accuracy and computational efficiency compared to the representative MSCKF-based (Multi-State Constraint Kalman Filter) open-source algorithm OpenVINS.