Abstract: To address map distortion caused by dynamic interference and loop closure failure under significant localization drift in simultaneous localization and mapping (SLAM), this study proposed a SLAM method integrating real-time dynamic point filtering and loop closure optimization. Based on the existing multi-sensor fusion SLAM framework, this study designed point cloud preprocessing and optimized loop detection. First, ground segmentation was implemented via grid feature analysis, and dynamic points were filtered using grid occupancy statistics to suppress motion interference and optimize SLAM mapping results. Subsequently, radius search was replaced by binary triangle combined descriptor (BTCD) matching for loop closure initialization. Geometric feature matching generated coarse pose estimation, which served as the initial value for the iterative closest point (ICP) algorithm, accelerating point cloud registration and improving robustness to optimize SLAM localization results. Experiments show that the proposed method can quickly eliminate dynamic interference of the map in dynamic scenes, reduce loop closure time, and improve robustness of SLAM system localization and the reliability of mapping.