AD-LIO, Robust Tightly Coupled LiDAR-inertial Odometry Based On The Adaptive Voxel Resolution Upgrade And Degeneracy Detection

This is the authors' abstract. We don't add key points for this paper.

This paper proposes a LiDAR-inertial odometry (LIO) based on the adaptive voxel resolution upgrade and degeneracy detection and optimization method, AD-LIO. This approach proposes an adaptive voxel resolution upgrade method (AVRU) to address the issue of fixed voxel size and inconsistent feature scales in natural environments. The AVRU module improves the accuracy of point cloud registration by performing voxel segment and feature-based clustering on point clouds in different regions. It divides the originally blurry voxel into smaller voxels with higher resolution, thereby enhancing the robustness and precision of the LIO system. In the meantime, this paper introduces a degeneracy detection method based on the Localization Parameter, and proposes a renewable optimization module to more strictly screen feature matching in degraded environments. Thus, it reduces the impact of noise on system robustness and improves the performance of LIO systems in harsh environments. The proposed method is extensively evaluated on datasets collected from various environments.