Multi-Sector Overlap Loss: A Universal Framework for One-Shot 6DoF Global Localization Across Heterogeneous LiDARs

Abstract

This paper presents a universal LiDAR point cloud global localization framework based on multi-sector overlapping loss to address the localization challenges caused by heterogeneous LiDAR point clouds with varying resolutions, scanning formats, and field of view differences. The proposed method first transforms point clouds into cylindrical coordinates and projects them by sectors onto the cylindrical radial-height plane, effectively encoding scene contour information. Feature descriptors are then generated through flattening and dimensionality reduction, combined with sliding window techniques and overlap reward-penalty strategies to enhance inter-sector distinctiveness, enabling accurate candidate queue generation and initial rotation estimation. Additionally, a voxel-overlap-based similarity detection mechanism filters potential false candidates, and the GICP algorithm is applied to obtain precise 6-DoF global pose estimation. Experimental evaluation on eight sequences across three heterogeneous point cloud scenarios in the HeLiPR dataset demonstrates that our method exhibits superior robustness and accuracy compared to state-of-the-art LiDAR global localization algorithms. Notably, the framework also performs excellently in homogeneous LiDAR global localization scenarios, proving its broad applicability.