Mesh-LOAM: Real-Time Mesh-Based LiDAR Odometry and Mapping

Although having achieved real-time performance on mesh construction, most of the existing LiDAR odometry and meshing methods have difficulties in dealing with cluttered scenes due to relying on explicit meshing techniques. To overcome these limitations, we propose a real-time mesh-based LiDAR odometry and mapping approach for large-scale scenes through implicit reconstruction and a parallel spatial-hashing scheme. In order to efficiently reconstruct the triangular meshes using the implicit function, we suggest an incremental voxel meshing strategy that depends on a novel voxel map fusing scans through a single traversal of the current frame. Moreover, we introduce a scalable partition module to compress space. By taking advantage of the rapid access to triangular meshes, we design a robust odometry method with location and feature-based data association to estimate the poses between the input point clouds and the recovered triangular meshes. The experimental results on four public datasets demonstrate the effectiveness of our proposed approach in recovering both accurate motion trajectories and environmental mesh maps.