DALO-SLAM: LiDAR-only SLAM based on dynamic removal and adaptive feature enhancement in GNSS-denial environment

Abstract In environments lacking global navigation satellite system (GNSS) signals, light detection and ranging (LiDAR)-only simultaneous localization and mapping (SLAM) is prone to accumulating pose estimation errors due to dynamic interference and low feature density, affecting system accuracy and stability. Achieving a balance between the robustness of traditional methods in dynamic environments and the effectiveness of pose constraints in sparse feature scenes remains challenging. To address this issue, this paper proposes a LiDAR-only SLAM method based on dynamic removal and adaptive feature enhancement, referred to as DALO-SLAM, aiming to improve system adaptability and accuracy in complex environments. First, the dynamic point cloud is efficiently removed during preprocessing by integrating graph-based invariant random sample consensus (GI-RANSAC). Second, an adaptive feature enhancement strategy is introduced, incorporating intensity information and local structural features in the degradation direction. Stability is further improved through adaptive weight adjustment based on degradation perception. The experiments on the KITTI dataset and real-world scenarios demonstrate that DALO-SLAM outperforms existing state-of-the-art methods in complex environments.