CTF-LIOM: Coarse-to-Fine Loop Closure Detection in Lidar-Inertial Odometry and Mapping

We propose CTF-LIOM, a novel coarse-to-fine loop closure detection method for lidar-inertial odometry and mapping, which significantly enhances global pose estimation consistency and mapping quality. The approach integrates structured steps: data preprocessing, loop detection using the Stable Triangle Descriptor (STD), coarse matching based on plane overlap, and subgraph fine matching with Nano-VGICP. This framework effectively identifies accurate loop frame indices and corresponding pose constraints. Experiments on the M2DGR door_02 dataset demonstrate that CTF-LIOM achieves an average error of just 0.19 meters with an RMSE of 0.20 meters, delivering a 47.1% improvement over Faster-LIO. On the street_04 dataset, incorporating backend loop detection and global pose optimization yields a 58.62% improvement in accuracy over CTF-LIO. Overall, CTF-LIOM exhibits superior trajectory accuracy and robustness in complex environments and during prolonged SLAM tasks, effectively reducing ghosting and map overlap.