QualiSLAM: Quality-Driven Feature Adaptation for Visual SLAM in Challenging Environments

It is well known that visual simultaneous localization and mapping (SLAM) is essential for autonomous systems navigating in unknown environments. However, conventional methods often degrade under challenging visual conditions, such as illumination changes, motion blur, and texture sparsity. The study of this article is aimed at proposing a novel quality-driven feature adaptation method, called QualiSLAM, to enhance the robustness and accuracy of visual SLAM. The proposed approach dynamically evaluates feature confidence using selected image quality metrics across different regions in the frame. This approach, together with an adaptive metrics selection strategy, can be shown to reduce redundancy and improve efficiency. Finally, a criteria importance through intercriteria correlation (CRITIC)-based weighting mechanism further integrates these metrics, thereby dynamically adjusting feature contributions during pose estimation. Based on public datasets and real-world experimental data, experiments demonstrate significant improvements in localization accuracy compared to current state-of-the-art methods, particularly in varying challenging scenarios. Importantly, QualiSLAM functions as an independent module with strong generalization capabilities, seamlessly integrates into both visual SLAM and visual-inertial navigation systems (VINSs) to enhance pose estimation performance.