NE-LOAM: A Point Cloud Normals Enhanced LiDAR Odometry

In recent years, with the rapid development of autonomous driving, SLAM (Simultaneous Localization and Mapping) technology has garnered widespread attention. This paper proposes a LiDAR odometry method called NE-LOAM, which is based on enhanced point cloud normals, aimed at addressing the drift issues faced by existing SLAM algorithms in noisy and feature-sparse complex outdoor environments. Firstly, NE-LOAM introduces an enhanced feature point selection method that combines curvature and normals to improve the accuracy of point cloud matching. Secondly, to address the increased computational burden of normal calculation, NE-LOAM employs an improved Fast Approximate Least Squares (FALS) method, called Beam Projection-based Fast Normal Estimation (BPFNE), to quickly compute normals. Thirdly, NE-LOAM presents a novel correspondence selection method based on geometric and normal consistency to eliminate incorrect matches in point cloud matching, thereby achieving more accurate pose estimation. Finally, experimental results on the KITTI and M2DGR datasets demonstrate that NE-LOAM achieves higher pose estimation accuracy and real-time performance in complex environments. Our proposed method is not only theoretically sound but also practically necessary, significantly enhancing the robustness and accuracy of the system.