A LiDAR SLAM Method with Adaptive Point Cloud Feature Extraction

Addressing the issue that traditional SLAM systems’ front-end odometry, which relies on fixed thresholds for feature point extraction, may lead to reduced accuracy in pose estimation, this paper proposes an adaptive point cloud feature extraction method. By introducing a non-linear attenuation coefficient, the method adaptively adjusts the feature extraction threshold based on the distance of the point cloud to the LiDAR center, while adding a non-linear weight to enhance the robustness of the system’s pose estimation. Based on the FAST-LIO2 algorithm, the reliability of the proposed method is validated using the M2DGR dataset. Experimental results demonstrate that the localization and mapping accuracy of the proposed method surpasses that of similar algorithms, ensuring operational efficiency while improving accuracy, thereby effectively enhancing the applicability of the algorithm.