A place recognition and localization method based on vehicle-mounted LiDAR bird’s-eye view (BEV) rotation-invariant features is proposed for autonomous positioning in urban road scenes. Firstly, three-dimensional point clouds acquired by vehicle-mounted LiDAR are transformed into two-dimensional BEV images through a normalized point-density representation. Subsequently, the translation and rotation invariance properties of LiDAR BEV features are analyzed. On this basis, a rotation-translation-equivariant local feature extraction network for BEV images, termed rotation equivariant local feature extraction network (RE-Net), is designed, together with a global feature aggregation strategy. Finally, place recognition and localization in urban road scenes are performed based on the rotation-invariant BEV features extracted from vehicle-mounted LiDAR data. The experimental results show that the proposed method improves Recall@1 on the NCLT dataset by at least 5 percentage points compared with mainstream methods, achieves a Recall@1 of over 90% on the self-collected urban wide-angle LiDAR dataset, and demonstrates that training data diversity contributes more than data scale to model generalization.
激光雷达能够主动获取车辆周围环境的三维点云数据,相比车载摄像机对光照变化、弱纹理或无纹理场景更具鲁棒性,在激光雷达地点识别(LiDAR Place Recognition, LPR)、全局定位(Global Localization)等任务中有独特优势[1]。但因激光点云存在稀疏、无序和非结构化的特点,如何对其进行有效表征是激光雷达地点识别面临的重要挑战[2]。
采用首位召回率(Recall@1)为地点识别的评估指标。若查询帧与Top-1匹配帧之间的距离在5 m以内,视为成功定位。为衡量描述符对正确匹配与错误匹配的区分能力,进一步采用最大F1分数(max F1 score),通过遍历描述符距离阈值,统计不同阈值下的精确率(Precision)、召回率(Recall)和F1分数,并取最大F1分数作为最终结果。
UyM A, LeeG H. PointNetVLAD: deep point cloud based retrieval for large-scale place recognition[C]//Proceedings of the 2018 IEEE/CVF conference on computer vision and pattern recognition. Piscataway, USA: IEEE, 2018:4470-4479.
[2]
YinH, XuX C, LuS, et al. A survey on global LiDAR localization: challenges, advances and open problems[J]. International Journal of Computer Vision, 2024,132(8):3139-3171.
[3]
LuoK, YuH S, ChenX, et al. 3D point cloud-based place recognition: a survey[J]. Artificial Intelligence Review, 2024,57(4):83.
[4]
XuX C, YinH, ChenZ X, et al. DiSCO: differentiable scan context with orientation[J]. IEEE Robotics and Automation Letters, 2021,6(2):2791-2798.
[5]
LuoL, CaoS Y, HanB, et al. BVMatch: LiDAR-based place recognition using bird’s-eye view images[J]. IEEE Robotics and Automation Letters, 2021,6(3):6076-6083.
[6]
LuoL, ZhengS H, LiY X, et al. BEVPlace: learning LiDAR-based place recognition using bird’s eye view images[C]//Proceedings of the 2023 IEEE/CVF international conference on computer vision. Piscataway, USA: IEEE, 2023: 8666-8675.
[7]
ArandjelovicR, GronatP, ToriiA, et al. NetVLAD: CNN architecture for weakly supervised place recognition[C]//Proceedings of the 2016 IEEE conference on computer vision and pattern recognition. Piscataway, USA: IEEE, 2016:5297-5307.
[8]
XuX C, LuS, WuJ, et al. RING: roto-translation invariant gram for global localization on a sparse scan map[J]. IEEE Transactions on Robotics, 2023,39(6):4616-4635.
[9]
ShenY Q, TunaT, HutterM, et al. ForestLPR: LiDAR place recognition in forests attentioning multiple BEV density images[C]//Proceedings of the 2025 IEEE/CVF conference on computer vision and pattern recognition. Piscataway, USA: IEEE, 2025:6659-6669.
[10]
WangZ Y, ShiC H, WangN, et al. BEVDiffLoc: end-to-end LiDAR global localization in BEV view based on diffusion model[C]//Proceedings of the 2025 IEEE/RSJ international conference on intelligent robots and systems. Piscataway, USA: IEEE, 2025:16240-16246.
[11]
ZhangC H, LuoL, CaoS Y, et al. S-BEVLoc: BEV-based self-supervised framework for large-scale LiDAR global localization[J]. IEEE Robotics and Automation Letters, 2025,10(10):9614-9621.
[12]
KnightsJ, VidanapathiranaK, RamezaniM, et al. Wild-places: a large-scale dataset for LiDAR place recognition in unstructured natural environments[C]//Proceedings of the 2023 IEEE international conference on robotics and automation. Piscataway, USA: IEEE, 2023:11322-11328.
[13]
LiY H, LiH. LiDAR-based initial global localization using two-dimensional (2D) submap projection image (SPI)[C]//Proceedings of the 2021 IEEE international conference on robotics and automation. Piscataway, USA: IEEE, 2021:5063-5068.
[14]
LecunY, BottouL, BengioY, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998,86(11):2278-2324.
LuoL, CaoS Y, LiX R, et al. BEVPlace++: fast, robust, and lightweight LiDAR global localization for autonomous ground vehicles[J]. IEEE Transactions on Robotics, 2025,41:4479-4498.
[17]
Carlevaris-BiancoN, UshaniA K, EusticeR M. University of Michigan North Campus long-term vision and lidar dataset[J]. The International Journal of Robotics Research, 2016,35(9):1023-1035.