PF-LIO: Tightly-Coupled Lidar-Inertial Odometry Based on Plane Fusion

In GNSS-denied environments, such as tunnels or buildings, low-beam light detection and ranging (LiDAR) systems struggle to constrain the z-axis due to insufficient vertical features. Without supplementary sensors, this limitation increases pitch angle and height estimation errors, causing z-axis errors to accumulate over time. To address these issues, this paper proposes a cost effective, robust, and lightweight LiDAR-inertial odometry (LIO) system based on a plane-fusion strategy. By integrating multi-modal sensor data, the system leverages environmental geometry and motion information for precise localization in GNSS-denied scenarios with limited vertical features. The system incorporates voxel down sampling and point clouds motion distortion correction to reduce computational complexity and enhance robustness. A spatial hash function maps point clouds to voxel grids, eliminating the need for K-Nearest Neighbor (KNN) searches and improving registration efficiency. Weighted least squares (WLS) optimize plane parameters, while reducing plane fitting from six to three degrees of freedom (DoF) minimizes memory consumption. The system introduces a novel plane-fusion method combining Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and union-find algorithms to improve plane-fitting accuracy and reduce map uncertainty. Additionally, a slope detection module, integrated with an Iterated Error-State Kalman Filter (IESKF), mitigates z-axis drift. Experiments show the system outperforms state-of-the-art methods in challenging scenarios, offering a highly accurate and economical navigation solution.