Advanced LiDAR-Inertial Odometry with Minimized Drift Using Rasterized Ground Point Cloud Constraints

3D LiDAR-based odometry is a well-established technique for the simultaneous localization and mapping (SLAM) of unmanned ground vehicles. Despite its widespread use, current LiDAR odometry methods often suffer from vertical drift, compromising the precision and consistency of the point cloud map. This issue negatively affects localization accuracy and mapping quality, particularly with low vertical resolution LiDAR systems. We propose an enhanced LiDAR-inertial odometry (LIO) method that leverages rasterized ground point cloud (GPC) constraints to mitigate vertical drift. Our approach divides the current frame's GPC and GPC submaps into grid-based maps, using the GPC within each grid to capture local ground features. This improves the adaptability and robustness of the algorithm to diverse environments. Additionally, we adopt average point matching within grid cells to improve the efficiency of GPC alignment. Experimental results validates that the rasterized GPC matching significantly improves computation efficiency. Our method comparison with four LIO methods: LIO-SAM, LINS, DLIO, FAST-LIO2 and two classic ground-constrained methods: LeGO-LOAM, HDL SLAM. In the public dataset with a 32-line LiDAR, our proposed method reduces the average absolute trajectory error (ATE) in translation by 10.06% and along the Z-axis by 21.09% compared to the LIO-SAM. Furthermore, in our private dataset using 16-line LiDAR platform, we observe a dramatic 45.7% reduction in translation error and a 46.8% reduction in Z-axis error compared to DLIO. Moreover, our methods achieves better results than FAST-LIO2 on particular scenarios. The results highlight the superior performance and robustness of our method, particularly on low vertical resolution LiDAR platforms.