Robust and Low-Complexity Feature Matching for UAV SLAM in Repetitive Environments

• Proposes a low-complexity feature matching framework to address perceptual aliasing in UAV SLAM. • Introduces uniqueness scores and motion-consistent geometric constraints for robust feature matching. • Develops two practical pipelines: VIO-guided matching and Instant stereo-based matching without inertial priors. • Achieves accuracy comparable to state-of-the-art methods while reducing computational cost by up to 10x. • Enables real-time deployment on resource-constrained micro-UAV platforms in GPS-denied environments. Perceptual aliasing remains a significant challenge for feature-based simultaneous localization and mapping (SLAM), especially in resource-constrained unmanned aerial vehicles (UAVs) operating in GPS-denied environments. This paper presents a low-complexity image-matching approach that improves correspondence reliability while maintaining computational efficiency. Feature matching is formulated as a linear assignment problem, where the cost matrix is designed to integrate descriptor similarity, feature distinctiveness estimation, and motion-consistent geometric constraints. Two practical pipelines are proposed: a VIO-guided strategy that incorporates visual-inertial priors, and an Instant Matching strategy that estimates pose directly from stereo observations when inertial information is unreliable. Experiments on 42 TartanAir sequences show that the proposed methods achieve comparable accuracy to state-of-the-art deep learning methods like SuperGlue while reducing computational cost by up to 10x and significantly lowering memory usage. These properties make the approach well suited for real-time deployment on micro-UAV platforms.