Keyframe Selection Strategy for Visual SLAM based on Fuzzy Analysis

Keyframe Selection is essential in Visual SLAM systems. In complex environments, existing keyframe selection strategies often generate robot trajectories with significant errors due to unsatisfactory alignment with actual scenes. This paper proposes a keyframe selection strategy for visual SLAM, modeling robots' motion trajectory parameters using fuzzy sets. After representing each uncertain trajectory characteristic parameter by a fuzzy function, we build a fuzzy implication reasoning model for filtering frames with distinct trajectory characteristics, achieving real-time selection of keyframes. In the experiments, we integrate the proposed selection strategy with mainstream monocular visual SLAM algorithms and validate the effectiveness of our strategy using the KITTI dataset.