<p>This paper proposes Dynamic YOLOv8-RTAB-Map SLAM (DYR-SLAM), an enhanced dynamic visual SLAM algorithm that leverages YOLOv8 and RTAB-Map to address the challenges of inaccurate mask coverage and small pixel region omission in existing deep learning-based visual SLAM systems. DYR-SLAM employs YOLOv8 to obtain semantic information and an initial mask, which is subsequently corrected using multi-frame depth information to ensure mask coverage consistency with dynamic objects. By integrating multi-frame point cloud and depth information in RTAB-Map, the mask accuracy is optimized, retaining more static information, while eliminating dynamic objects. Furthermore, an acceleration constraint model predicts dynamic object trajectories to ensure robustness in complex motion and occlusion scenarios. Adaptive culling weights and multi-frame consistency detection improve point cloud culling accuracy when RTAB-Map is combined with YOLOv8. Evaluated on the TUM RGB-D dataset, DYR-SLAM demonstrates higher localization accuracy and better dense point cloud mapping in dynamic scenes compared to advanced SLAM systems like RTAB-MAP, DS-SLAM, and DynaSLAM.</p>