A LiDAR-Inertial Odometry by Fusing Adjacent Frames with Point Selection

LiDAR (Light Detection and Ranging) is the most commonly used sensor for autonomous driving. Spinning LiDAR is fabricated by stacking laser emitters and receivers in the vertical direction to form a 3D LiDAR, resulting in a low vertical resolution. Geometric features may be lost during scanning. The issue will be worsened in outdoor situations where the LiDAR is measuring over long distances. To address this issue, we propose a novel LiDAR-inertial Odometry (LiO) system that utilizes two frames of point cloud simultaneously, providing additional geometric information. Firstly, the previous frame’s point cloud is evaluated to determine whether it should be used, based on a set velocity threshold. Then, two frames are projected onto the same momentary coordinate system for point cloud de-distortion. To reduce the computational burden of processing two frames simultaneously, we have adopted a point selection approach. During the process of optimizing iterations, the points used to calculate residuals and construct maps are selected to achieve a balance between accuracy and efficiency. The method proposed in this study is based on the LIO-SAM implementation and has been tested on the KITTI and M2DGR datasets. The experimental results demonstrate that the proposed method enhances the localization accuracy and reduces the run time.