A-MSCKF: An Improved MSCKF Algorithm Based on Adaptive Threshold Feature Points

When implementing robot localization using the traditional Multi-State Constraint Kalman Filter (MSCKF) algorithm, the frontend relies on optical flow methods to track feature points. However, this approach results in low-precision feature point matching and lacks robustness to changes in environmental lighting conditions. Considering the complex lighting variations in different environments during localization and tracking, this paper proposes the A-MSCKF algorithm. The frontend of this algorithm employs an adaptive threshold ORB (Oriented FAST and Rotated BRIEF) feature point extraction method, and the backend of the MSCKF algorithm is optimized. Experimental results show that the proposed improved method exhibits better stability and lower tracking errors compared to traditional approaches in trajectory tracking comparisons.