Adaptive Multi-Scale Three-Dimensional Point Cloud Registration Network for Autonomous Driving

WANG Lin; HAN Dong; HAN Jingyi; CHEN Shengmei; LIU Cheng

doi:10.7652/xjtuxb202610003

您当前的位置：

首页 >

文章列表页 >

Adaptive Multi-Scale Three-Dimensional Point Cloud Registration Network for Autonomous Driving

Special Topic Fusion Perception and Decision-Making Planning for Intelligent Driving | 更新时间：2026-09-29

- Adaptive Multi-Scale Three-Dimensional Point Cloud Registration Network for Autonomous Driving
- Journal of Xi'an Jiaotong University Vol. 60, Issue 10, Pages: 34-44(2026)
- 作者机构：
  
  西北大学电子信息学院，710127，西安
- 作者简介：
- 基金信息：
- DOI：10.7652/xjtuxb202610003
  CLC： TP29
- Received：21 April 2026，
  
  Online First：24 June 2026，
  
  Published：10 October 2026
- 稿件说明：
移动端阅览
WANG Lin, HAN Dong, HAN Jingyi, et al. Adaptive Multi-Scale Three-Dimensional Point Cloud Registration Network for Autonomous Driving[J/OL]. Journal of Xi'an Jiaotong University,2026,60 (10):34-44. https://doi.org/10.7652/xjtuxb202610003.
DOI：

WANG Lin, HAN Dong, HAN Jingyi, et al. Adaptive Multi-Scale Three-Dimensional Point Cloud Registration Network for Autonomous Driving[J/OL]. Journal of Xi'an Jiaotong University,2026,60 (10):34-44. https://doi.org/10.7652/xjtuxb202610003. DOI：

摘要

针对智能驾驶场景中，激光雷达点云稀疏、特征模糊以及低重叠引发关键点误匹配的问题，提出了一种面向智能驾驶的自适应多尺度三维点云配准网络。首先，该网络利用自适应多尺度特征提取模块获取采样点的多层级特征，并结合自适应邻域稳定器实现关键点的精确定位，从而解决由于特征模糊而导致的匹配难题。其次，通过融合多尺度特征匹配一致性与几何一致性校验，提升待配准点对的内点置信度。最后，基于该置信度，采用由粗到精的迭代优化策略，完成三维点云的精确配准。在KITTI数据集上的测试结果表明：该网络的相对平移误差和相对旋转误差分别低至0.0618m和0.21°，配准召回率高达99.8%。在nuScenes数据集上的测试结果表明：在帧间距离分别为10、20、30m时，配准召回率分别达到90.1%、83.3%和78.8%。上述结果验证了该网络能够有效抑制复杂道路场景下的误匹配，实现高精度、高鲁棒性的大规模三维点云配准。该研究适配了动态复杂路况下的实际应用需求，可为自动驾驶场景下的精准定位、环境感知提供可靠的技术支撑。

Abstract

An adaptive multi-scale three-dimensional point cloud registration network for autonomous driving is proposed to address the issues of sparse light detection and ranging point clouds

ambiguous features

and keypoint mismatches caused by low overlap in autonomous driving scenarios. First

multi-level features of sampled points are extracted via an adaptive multi-scale feature extraction module

and precise keypoint localization is achieved by incorporating an adaptive neighborhood stabilizer

thereby resolving the matching difficulties caused by feature ambiguity. Second

the inlier confidence of the point pairs to be registered is enhanced by integrating multi-scale feature matching consistency with geometric consistency verification. Finally

based on this confidence

a coarse-to-fine iterative optimization strategy is adopted to accomplish accurate three-dimensional point cloud registration. Test results on the KITTI dataset indicate that a relative translation error as low as 0.0618mand a relative rotation error as low as 0.21° are achieved by the network

with a registration recall of up to 99.8%. Furthermore

test results on the nuScenes dataset show that registration recalls of 90.1%

83.3%

and 78.8% are achieved at inter-frame distances of 10

and 30m

respectively. It is validated by the aforementioned results that mismatches in complex road scenarios are effectively suppressed by the proposed network

and high-precision

highly robust large-scale three-dimensional point cloud registration is realized. The practical application requirements under dynamic and complex traffic conditions are met by this research

and reliable technical support is provided for precise localization and environmental perception in autonomous driving scenarios.

关键词

Keywords

references

Yu Zhao, Zhang Peize, Shi Jing.Transformation of industrial robotics with natural language models: recent progress and future prospects [J].Robotics and Computer-Integrated Manufacturing, 2026, 97: 103113.

周翔, 宋林烨, 夏海峰, 等.个性化热舒适系统在电动汽车中的应用与展望[J].西安交通大学学报, 2026, 60(1): 1-13.

Zhou Xiang, Song Linye, Xia Haifeng, et al. Application and prospects of personalized thermal comfort systems in electric vehicles[J].Journal of Xi'an Jiaotong University, 2026, 60(1): 1-13.

Song C H, Blukis V, Tremblay J, et al. Robospatial: teaching spatial understanding to 2D and 3D vision-language models for robotics [C]//2025IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR ). Piscataway, NJ, USA: IEEE, 2025: 15768-15780.

Wang Zhigang, Su Yifei, Li Chenhui, et al. Openvocabulary octree-graph for 3D scene understanding [C]//2025IEEE/CVF International Conference on Computer Vision (ICCV).Piscataway, NJ, USA: IEEE, 2025: 7037-7047.

马庆禄, 白锋, 张杰, 等.基于激光雷达点云地图的车辆定位与导航[J].光学精密工程, 2024, 32(16): 2537-2549.

Ma Qinglu, Bai Feng, Zhang Jie, et al. Vehicle localization and navigation method based on LiDAR point cloud map[J]. Optics and Precision Engineering, 2024, 32(16): 2537-2549.

Afzal Maken F, Muthu S, Nguyen C, et al. Improving 3D reconstruction through RGB-D sensor noise modeling [J].Sensors, 2025, 25(3): 950.

Tian Yifei, Li Xiangyu, Yin Jieming.From pseudo-to non-correspondences: robust point cloud registration via thickness-guided self-correction [J].Computers &Graphics, 2026, 134: 104528.

Ma Kaifeng, Yan Fengtao, Li Shiming, et al. Lowoverlap registration of multi-source LiDAR point clouds in urban scenes through dual-stage feature pruning and progressive hierarchical methods [J]. Remote Sensing, 2025, 17(17): 2938.

Kashef Tabrizian S, Terryn S, Vanderborght B.Toward autonomous self-healing in soft robotics: a review and perspective for future research [J].Advanced Intelligent Systems, 2025, 7(8): 2400790.

王丞, 田暄, 郭瑞, 等.自适应Harris角点提取的点云粗配准算法[J].西安交通大学学报, 2022, 56(3): 33-44.

Wang Cheng, Tian Xuan, Guo Rui, et al. Coarse point cloud registration based on adaptive Harris corner extraction[J].Journal of Xi'an Jiaotong University, 2022, 56(3): 33-44.

Wang Weijie, Mei Guofeng, Zhang Jian, et al. Fullygeometric cross-attention for point cloud registration [C]//2025International Conference on 3D Vision (3DV). Piscataway, NJ, USA: IEEE, 2025: 347-356.

Xia Yaqi, Xia Yan, Li Wei, et al. ASFM-Net: asymmetrical Siamese feature matching network for point completion [C]//Proceedings of the 29th ACM International Conference on Multimedia.New York, NY, USA: ACM, 2021: 1938-1947.

Besl P J, Mckay N D.A method for registration of 3-D shapes [J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 1992, 14(2): 239-256.

Wang Yue, Solomon J.Deep closest point: learning representations for point cloud registration [C]//2019IEEE/CVF International Conference on Computer Vision (ICCV).Piscataway, NJ, USA: IEEE, 2019: 3522-3531.

Yew Z J, Lee G H.RPM-Net: robust point matching using learned features [C]//2020IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2020: 11821-11830.

Fu Kexue, Liu Shaolei, Luo Xiaoyuan, et al. Robust point cloud registration framework based on deep graph matching [C]//2021IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2021: 8889-8898.

Bai Xuyang, Luo Zixin, Zhou Lei, et al. PointDSC: robust point cloud registration using deep spatial consistency [C]//2021IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2021: 15854-15864.

Yew Z J, Lee G H.REGTR: end-to-end point cloud correspondences with transformers [C]//2022IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2022: 6667-6676.

Qin Zheng, Yu Hao, Wang C, et al. Geometric transformer for fast and robust point cloud registration [C]//2022IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2022: 11133-11142.

Qi C R, Yi Li, Su Hao, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space [C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., 2017: 5105-5114.

Geiger A.Are we ready for autonomous driving? The KITTI vision benchmark suite [C]//Proceedings of the 2012IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2012: 3354-3361.

Choy C, Park J, Koltun V, et al. Fully convolutional geometric features [C]//2019IEEE/CVF International Conference on Computer Vision (ICCV).Piscataway, NJ, USA: IEEE, 2019: 8957-8965.

Yu H, Li F, Saleh M, et al. CoFiNet: reliable coarse-to-fine correspondences for robust point cloud registration [J]. Advances in Neural Information Processing Systems, 2021, 34: 23872-23884.

Huang Shengyu, Gojcic Z, Usvyatsov M, et al. PREDATOR: registration of 3D point clouds with low overlap [C]//2021IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2021: 4265-4274.

Yan Shaocheng, Shi Pengcheng, Zhao Zhenjun, et al. Turboreg: turboclique for robust and efficient point cloud registration [C]//2025IEEE/CVF International Conference on Computer Vision (ICCV).Piscataway, NJ, USA: IEEE, 2025: 26371-26381.

Caesar H, Bankiti V, Lang A H, et al. nuScenes: a multimodal dataset for autonomous driving [C]//2020IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ, USA: IEEE, 2020: 11618-11628.

Views

112

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

A Generalized Zero-Shot Driving Behavior Recognition Method for Intelligent Driving

Research and Practice on Evolutionary Networks

Adaptive Frequency-Shifted and Re-Scaling Stochastic Resonance with Applications to Fault Diagnosis

A Fast Motion Estimation Algorithm for Motion Complexity Decision

Drag Reduction Characteristics of Longitudinal Grooves under Limited Gap Height Conditions

Related Author

HAN Jinyi

XIE Jiahao

MENG Xianjia

XU Zuyin

许大炜 1

黄泳翔 1

栾钟治 2

王锐 2

Related Institution

College of Computer Science, Northwest University

School of Electronics and Information Engineering, Xi'an Jiaotong University

State Key Laboratory for Manufacturing Systems Engineering, Xi'an Jiaotong University

College of Information Science and Engineering, Huaqiao Univesity

Xiamen Key Laboratory of Application-Specific Integrated Circuit System

⁰