Fusion SLAM with adaptive keyframe selection and LSD and LBD line features

This study addresses the challenges of reduced robustness in complex environments and unstable feature extraction in lowtexture scenes commonly encountered in visual SLAM. An improved system is proposed with two main innovations: (1) a camera geometry-constrained adaptive keyframe selection algorithm, which dynamically adjusts the insertion strategy according to scene complexity and camera motion, thereby enhancing adaptability and robustness in challenging environments; (2) the integration of LSD line detection with LBD descriptors to extract stable line features from static regions, compensating for the limitations of point features in texture-sparse scenes and improving overall mapping accuracy. Experimental results verify that the proposed system provides improved reliability and estimation accuracy in both dynamic scenes and low-texture environments.