A Multiview LiDAR Measurement for Robust Wheel Localization in Low-Overlap Smart Garage Environments

Abstract

Accurate localization of wheel centers is essential for autonomous parking and robotic grasping in smart garages, yet low-overlap views and occlusions significantly degrade point cloud quality and measurement reliability. To address these challenges, this article proposes a multiview LiDAR-based measurement method for wheel localization in smart garage environments, providing a complete perception pipeline from raw point cloud acquisition to wheel-center estimation. First, an overlap-aware registration strategy integrating ground-plane constraints and adaptive voxel weighting is introduced to improve the accuracy of multiview point cloud alignment under low-overlap conditions. Second, a multimodal hierarchical attention network is proposed to effectively model sparse and boundary features, enabling precise wheel segmentation. Subsequently, a point cloud completion method based on local surface fitting and normal-guided neighborhood interpolation is proposed to recover occlusion-induced structural deficiencies and improve geometric integrity. Experiments conducted in real garage environments demonstrate that the proposed method achieves a rotation error (RE) of 0.96°, a translation root-mean-square error (RMSE) of 8.3 mm, and a segmentation mIoU of 89.8 %, validating its effectiveness for reliable wheel localization under multiview, low-overlap constraints.