Page 371 - 《软件学报》2025年第12期

5752 软件学报 2025 年第 36 卷第 12 期

表 12 不同方法的计算复杂度 (续)

方法 参数量 (G) FLOPs (M) 训练时间 (s) 测试时间 (s)
OGMM 10.09 4.58 0.052 5 0.027 6
RGM 51.83 24.46 0.102 1 0.043 2
GeoTransformer － 4.49 0.069 4 0.032 8
IFNet 32.63 5.50 0.096 9 0.065 5
GNN-GSSC 53.98 19.20 0.103 8 0.044 8
本文方法 26.22 13.21 0.109 8 0.037 1

3 总 结

本文提出了一种局部-全局动态图学习与互补融合的点云配准方法, 通过结合动态偏移的局部图学习、动态
关注的全局图学习和注意力驱动的互补融合模块, 有效地提升了点云配准的精度. 在特征提取阶段, 根据动态偏移
得到的代理点来学习特征空间的图结构, 获得了更具判别性和细节增强的局部特征表示. 同时, 利用特征学习的自
适应阈值动态地构建全局图, 进一步实现了点之间的长程依赖关系. 在特征匹配阶段, 注意力驱动的互补融合模块
通过交叉注意力机制提取相似和差异信息, 并采用自注意力机制进一步优化了特征的关联性. 实验结果表明, 本文
方法取得了最优的配准效果, 同时保持了较高的计算效率. 在未来的研究中, 可以针对不同场景进一步优化算法,
以提高其在复杂环境中的鲁棒性, 特别是对高噪声与部分遮挡点云的处理.

References:
[1] Dang Z, Wang LZ, Guo Y, Salzmann M. Match normalization: Learning-based point cloud registration for 6D object pose estimation in
the real world. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2024, 46(6): 4489–4503. [doi: 10.1109/TPAMI.2024.3355198]
[2] Jiang HB, Salzmann M, Dang Z, Xie J, Yang J. SE(3) diffusion model-based point cloud registration for robust 6D object pose
estimation. In: Proc. of the 37th Int’l Conf. on Neural Information Processing Systems. New Orleans: Curran Associates Inc., 2023.
21285–21297.
[3] Shi CH, Chen XYL, Lu HM, Deng WB, Xiao JH, Dai B. RDMNet: Reliable dense matching based point cloud registration for
autonomous driving. IEEE Trans. on Intelligent Transportation Systems, 2023, 24(10): 11372–11383. [doi: 10.1109/TITS.2023.3286464]
[4] Zhou XY, Lin ZW, Shan XJ, Wang YT, Sun DQ, Yang MH. DrivingGaussian: Composite Gaussian splatting for surrounding dynamic
autonomous driving scenes. In: Proc. of the 2024 IEEE/CVF Conf. on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024.
21634–21643. [doi: 10.1109/CVPR52733.2024.02044]
[5] Li GF, Fan HW, Jiang GZ, Jiang D, Liu YT, Tao B, Yun JT. RGBD-SLAM based on object detection with two-stream YOLOv4-
MobileNetv3 in autonomous driving. IEEE Trans. on Intelligent Transportation Systems, 2024, 25(3): 2847–2857. [doi: 10.1109/TITS.
2023.3284228]
[6] Wang YN, Tian YB, Chen JW, Xu K, Ding XL. A survey of visual SLAM in dynamic environment: The evolution from geometric to
semantic approaches. IEEE Trans. on Instrumentation and Measurement, 2024, 73: 2523221. [doi: 10.1109/TIM.2024.3420374]
[7] Besl PJ, McKay ND. A method for registration of 3-D shapes. IEEE Trans. on Pattern Analysis & Machine Intelligence, 1992, 14(2):
239–256. [doi: 10.1109/34.121791]
[8] Segal A, Haehnel D, Thrun S. Generalized-ICP. In: Proc. of Robotics: Science and Systems. 2009. [doi: 10.15607/RSS.2009.V.021]
[9] Yang JL, Li HD, Jia YD. Go-ICP: Solving 3D registration efficiently and globally optimally. In: Proc. of the 2013 IEEE Int’l Conf. on
Computer Vision. Sydney: IEEE, 2013. 1457–1464. [doi: 10.1109/ICCV.2013.184]
[10] Magnusson M, Lilienthal A, Duckett T. Scan registration for autonomous mining vehicles using 3D-NDT. Journal of Field Robotics,
2007, 24(10): 803–827. [doi: 10.1002/rob.20204]
[11] Aiger D, Mitra NJ, Cohen-Or D. 4-points congruent sets for robust pairwise surface registration. ACM Trans. on Graphics, 2008, 27(3):
1–10. [doi: 10.1145/1360612.1360684]
[12] Aoki Y, Goforth H, Srivatsan RA, Lucey S. PointNetLK: Robust & efficient point cloud registration using PointNet. In: Proc. of the 2019
IEEE/CVF Conf. on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019. 7156–7165. [doi: 10.1109/CVPR.2019.00733]
[13] Qi CR, Su H, Mo KC, Guibas LJ. PointNet: Deep learning on point sets for 3D classification and segmentation. In: Proc. of the 2017
IEEE Conf. on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017. 652–660. [doi: 10.1109/CVPR.2017.16]