SE-LIO: Semantic-Enhanced Solid-State-LiDAR–Inertial Odometry for Tree-Rich Environments

LiDAR-inertial odometry (LIO) based on line and plane features will face great challenges in environments abundant with unstructured features, such as tree-rich environments. In this article, we propose a semantic-enhanced solid-state-LIO (SE-LIO) in tree-rich environments. Multiple LiDAR frames are first merged and compensated with the inertial navigation system (INS) to increase the point-cloud coverage, thus improving the accuracy of semantic segmentation. Then, the unstructured point clouds, such as tree leaves and dynamic objects, are removed with the semantic information. Besides, the pole-like point clouds, primarily tree trunks, are modeled as cylinders to improve positioning accuracy. We also propose an adaptive piecewise cylinder-fitting method to accommodate environments with a high prevalence of curved tree trunks. Finally, an efficient iterated error-state Kalman filter (IESKF) is employed to integrate LiDAR and inertial measurements tightly. Exhaustive experiments are conducted on public and private datasets with complex campus and park scenes to evaluate the proposed SE-LIO. The results demonstrate that the proposed baseline LIO yields improved or comparable accuracy to the state-of-the-art methods. By incorporating the semantic enhancement, the absolute translation accuracy of the proposed SE-LIO on the public and private datasets is improved by 17.1% and 43.1%, respectively, compared to the baseline system. Besides, the ablation experiment results illustrate that the positioning accuracy is also improved by employing the proposed adaptive cylinder fitting.