广西师范大学学报（自然科学版） ›› 2024, Vol. 42 ›› Issue (3): 108-120.doi: 10.16088/j.issn.1001-6600.2023082502

• 研究论文 • 上一篇    下一篇

多尺度特征融合的点云配准算法研究

易见兵*, 彭鑫, 曹锋, 李俊, 谢唯嘉   

  1. 江西理工大学 信息工程学院, 江西 赣州 341000
  • 收稿日期:2023-08-25 修回日期:2023-12-17 发布日期:2024-05-31
  • 通讯作者: 易见兵(1980—), 男, 江西宜春人, 江西理工大学副教授, 博士。E-mail: yijianbing8@jxust.edu.cn
  • 基金资助:
    国家自然科学基金(62066018, 72261018); 江西省自然科学基金(20181BAB202004); 江西省教育厅科技项目(GJJ210828, GJJ200818, GJJ180482); 江西省赣州市科技计划项目; 江西省研究生创新专项(YC2022-S640)

Research on Point Cloud Registration Algorithm Based on Multi-scale Feature Fusion

YI Jianbing*, PENG Xin, CAO Feng, LI Jun, XIE Weijia   

  1. College of Information Engineering, Jiangxi University of Science and Technology, Ganzhou Jiangxi 341000, China
  • Received:2023-08-25 Revised:2023-12-17 Published:2024-05-31

PDF (PC)

158

摘要： 现有点云配准算法提取的特征不够丰富,导致配准精度很难进一步提升。针对该问题,本文提出一种基于深度学习的多尺度特征融合点云配准算法。首先,利用EdgeConv提取多个不同尺度的特征,该特征能够保持局部几何结构特性;接着,引入非线性极化注意力对其输出特征进行筛选,从而提高特征信息的有效性;然后,将以上多尺度特征进行融合并再次利用EdgeConv提取其特征,从而提高特征的表达能力;在刚体姿态估计阶段,采用线性李代数处理旋转变换以充分挖掘点云中的变换信息;最后,根据配准过程中提取点云特征的变化,动态调整损失函数各组成部分的权重,获得更准确的模型预测结果。在ModelNet40数据集上进行实验,本文算法在训练集和测试集样本种类相同时的旋转误差为1.826 7,位移误差为0.001 0;在训练集和测试集的样本种类不相同时(泛化实验)的旋转误差为2.979 4,位移误差为0.001 0。实验结果表明,本文算法的配准精度相比当前主流算法均有提高且泛化性能较好。

关键词: 深度学习, 点云配准, 特征提取, 刚体目标, 姿态估计, 李代数

Abstract: The features extracted by the existing point cloud registration algorithms are not so rich, which makes it difficult to further improve the accuracy of the registration. To address this problem, a deep learning-based multi-scale feature fusion point cloud registration algorithm is proposed. EdgeConv is employed to extract multiple features of different scales through the algorithm at first, which can maintain the local geometric structure characteristics. Then Non-linear Polarized Self-attention is introduced to filter its output features, and thus the effectiveness of feature information is improved. And later the above multi-scale features are fused and EdgeConv is employed again to extract their features, thereby improving the expression ability of the features. In the rigid pose estimation stage, Lie algebra is used to process the rotational transformation to fully exploit the transformation information of the point cloud. According to the changes of the extracted point cloud features during the registration process, the weight values of the components of the loss function are dynamically adjusted to evaluate the prediction results of the model more accurately. Tested on the ModelNet40 dataset, when the sample types of the train and test sets are the same, the rotation error of the proposed algorithm is 1.826 7 and the displacement error is 0.001 0, and when the sample types of the train and test sets are not the same (experiments on generalization), the rotation error of the proposed algorithm is 2.979 4 and the displacement error is 0.001 0. The experimental results show that the registration accuracy of the proposed algorithm has improved compared with the current mainstream algorithms, and it exhibits good generalization performance.

Key words: deep learning, point cloud registration, feature extraction, rigid object, pose estimation, Lie algebra

中图分类号:  TP391.41

[1] 栾佳宁, 张伟, 孙伟, 等. 基于二维码视觉与激光雷达融合的高精度定位算法[J]. 计算机应用, 2021, 41(5): 1484-1491. DOI: 10.11772/j.issn.1001-9081.2020081162.
[2] 王任栋, 徐友春, 齐尧, 等. 一种鲁棒的城市复杂动态场景点云配准方法[J]. 机器人, 2018, 40(3): 257-265. DOI: 10.13973/j.cnki.robot.170429.
[3] 李美佳, 于泽宽, 刘晓, 等. 点云算法在医学领域的研究进展[J]. 中国图象图形学报, 2020, 25(10): 2013-2023. DOI: 10.11834/jig.200253.
[4] HUANG X S, MEI G F, ZHANG J, et al. A comprehensive survey on point cloud registration[EB/OL]. (2021-03-05)[2023-08-25]. https://arxiv.org/abs/2103.02690. DOI: 10.48550/arXiv.2103.02690.
[5] 阎翔鑫, 蒋峥, 刘斌. 基于角度约束的跨源点云配准算法[J]. 激光与光电子学进展, 2023, 60(22): 2215004. DOI: 10.3788/LOP230478.
[6] 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. DOI: 10.1109/34.121791.
[7] LIU H K, ZHANG Y, LEI L J, et al. Hierarchical optimization of 3D point cloud registration[J]. Sensors, 2020, 20(23): 6999. DOI: 10.3390/s20236999.
[8] 朱玉梅, 邢明义, 蔡静. 基于法向量权重改进的ICP算法[J]. 计量学报, 2023, 44(6): 852-857. DOI: 10.3969/j.issn.1000-1158.2023.06.03.
[9] 李茂月, 田帅, 刘硕, 等. 基于结构光在机测量的变形薄壁件点云配准方法[J]. 光电子·激光, 2022, 33(11): 1148-1157. DOI: 10.16136/j.joel.2022.11.0052.
[10] YEW Z J, LEE G H. RPM-Net: robust point matching using learned features[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2020: 11821-11830. DOI: 10.1109/CVPR42600.2020.01184.
[11] 张文丽, 程兰, 任密蜂, 等. 基于AGConv局部特征描述符的点云配准[J]. 计算机工程, 2023, 49(11): 231-237. DOI: 10.19678/j.issn.1000-3428.0066359.
[12] WANG Y, SOLOMON J M. Deep closest point: learning representations for point cloud registration[C]// 2019 IEEE/CVF International Conference on Computer Vision (ICCV). Los Alamitos, CA: IEEE Computer Society, 2019: 3522-3531. DOI: 10.1109/ICCV.2019.00362.
[13] 李健, 黄硕文, 冯凯, 等. 核相关神经网络点云自动配准算法[J]. 同济大学学报(自然科学版), 2022, 50(11): 1685-1692. DOI: 10.11908/j.issn.0253-374x.21319.
[14] 刘磊, 熊风光, 尹宇慧, 等. 多特征提取与匹配矩阵驱动的点云配准[J]. 计算机工程与设计, 2023, 44(5): 1419-1426. DOI: 10.16208/j.issn.1000-7024.2023.05.019.
[15] QI C R, SU H, MO K C, et al. PointNet: deep learning on point sets for 3D classification and segmentation[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2017: 77-85. DOI: 10.1109/CVPR.2017.16.
[16] QI C R, YI L, SU H, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space[C]// Advances in Neural Information Processing Systems 30 (NIPS 2017). Red Hook, NY: Curran Associates Inc., 2017: 5100-5109.
[17] AOKI Y, GOFORTH H, SRIVATSAN R A, et al. PointNetLK: robust & efficient point cloud registration using PointNet[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2019: 7156-7165. DOI: 10.1109/CVPR.2019.00733.
[18] LI X Q, PONTES J K, LUCEY S. PointNetLK revisited[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2021: 12758-12767. DOI: 10.1109/CVPR46437.2021.01257.
[19] HUANG X S, MEI G F, ZHANG J. Feature-metric registration: a fast semi-supervised approach for robust point cloud registration without correspondences[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2020: 11363-11371. DOI: 10.1109/CVPR42600.2020.01138.
[20] SARODE V, LI X Q, GOFORTH H, et al. PCRNet: point cloud registration network using PointNet encoding[EB/OL]. (2019-11-04)[2023-08-25]. https://arxiv.org/abs/1908.07906. DOI: 10.48550/arXiv.1908.07906.
[21] 武越, 苑咏哲, 岳铭煜, 等. 点云配准中多维度信息融合的特征挖掘方法[J]. 计算机研究与发展, 2022, 59(8): 1732-1741. DOI: 10.7544/issn1000-1239.20220042.
[22] WANG Y, SUN Y B, LIU Z W, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 146. DOI: 10.1145/3326362.
[23] LIU H J, LIU F Q, FAN X Y, et al. Polarized self-attention: towards high-quality pixel-wise regression[EB/OL]. (2021-07-08)[2023-08-25]. https://arxiv.org/abs/2107.00782. DOI: 10.48550/arXiv.2107.00782.
[24] SHEN W, ZHANG B B, HUANG S K, et al. 3D-rotation-equivariant quaternion neural networks[C]// Computer Vision-ECCV 2020: LNCS Volume 12365. Cham: Springer, 2020: 531-547. DOI: 10.1007/978-3-030-58565-5_32.
[25] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// Computer Vision-ECCV 2018: LNCS Volume 11211. Cham: Springer, 2018: 3-19. DOI: 10.1007/978-3-030-01234-2_1.
[26] YANG J L, LI H D, CAMPBELL D, et al. Go-ICP: a globally optimal solution to 3D ICP point-set registration[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(11): 2241-2254. DOI: 10.1109/TPAMI.2015.2513405.
[27] 秦庭威, 赵鹏程, 秦品乐, 等. 基于残差注意力机制的点云配准算法[J]. 计算机应用, 2022, 42(7): 2184-2191. DOI: 10.11772/j.issn.1001-9081.2021071319.
[28] 梁正友, 姚强, 孙宇, 等.基于多尺度特征融合和残差混合注意力的点云配准算法[J]. 计算机工程与设计, 2023, 44(9): 2650-2656. DOI: 10.16208/j.issn1000-7024.2023.09.012.
[29] KUROBE A, SEKIKAWA Y, ISHIKAWA K, et al. CorsNet: 3D point cloud registration by deep neural network[J]. IEEE Robotics and Automation Letters, 2020, 5(3): 3960-3966. DOI: 10.1109/LRA.2020.2970946.
[30] YI R B, LI J L, LUO L, et al. DOPNet: achieving accurate and efficient point cloud registration based on deep learning and multi-level features[J]. Sensors, 2022, 22(21): 8217. DOI: 10.3390/s22218217.
[31] SONG Y N, SHEN W M, PENG K K. A novel partial point cloud registration method based on graph attention network[J]. The Visual Computer, 2023, 39(3): 1109-1120. DOI: 10.1007/s00371-021-02391-0.
[32] RUSU R B, BLODOW N, BEETZ M. Fast point feature histograms (FPFH) for 3D registration[C]// 2009 IEEE International Conference on Robotics and Automation. Los Alamitos, CA: IEEE Computer Society, 2009: 3212-3217. DOI: 10.1109/ROBOT.2009.5152473.
[33] ZHANG J Y, YAO Y X, DENG B L. Fast and robust iterative closest point[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(7): 3450-3466. DOI: 10.1109/TPAMI.2021.3054619.
[34] LV C L, LIN W S, ZHAO B Q. KSS-ICP: point cloud registration based on kendall shape space[J]. IEEE Transactions on Image Processing, 2023, 32: 1681-1693. DOI: 10.1109/TIP.2023.3251021.
[1] 田晟, 胡啸. 基于Transformer模型的车辆轨迹预测[J]. 广西师范大学学报（自然科学版）, 2024, 42(3): 47-58.
[2] 肖宇庭, 吕晓琪, 谷宇, 刘传强. 基于拆分残差网络的糖尿病视网膜病变分类[J]. 广西师范大学学报（自然科学版）, 2024, 42(1): 91-101.
[3] 高飞, 郭晓斌, 袁冬芳, 曹富军. 改进PINNs方法求解边界层对流占优扩散方程[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 33-50.
[4] 林玩聪, 韩明杰, 靳婷. 基于数据增强的多层次论点立场分类方法[J]. 广西师范大学学报（自然科学版）, 2023, 41(6): 62-69.
[5] 蒋懿波, 刘会家, 吴田. 基于改进残差网络的输电线路雷击过电压识别研究[J]. 广西师范大学学报（自然科学版）, 2023, 41(4): 74-83.
[6] 梁镇锋, 夏海英. 一种面向无人机航拍图像的快速拼接算法[J]. 广西师范大学学报（自然科学版）, 2023, 41(3): 41-52.
[7] 杨烁祯, 张珑, 王建华, 张恒远. 声音事件检测综述[J]. 广西师范大学学报（自然科学版）, 2023, 41(2): 1-18.
[8] 王鲁娜, 杜洪波, 朱立军. 基于流形正则的堆叠胶囊自编码器优化算法[J]. 广西师范大学学报（自然科学版）, 2023, 41(2): 76-85.
[9] 于梦竹, 唐振军. 基于手工特征的视频哈希研究综述[J]. 广西师范大学学报（自然科学版）, 2022, 40(5): 72-89.
[10] 张萍, 徐巧枝. 基于多感受野与分组混合注意力机制的肺结节分割研究[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 76-87.
[11] 李永杰, 周桂红, 刘博. 基于YOLOv3模型的人脸检测与头部姿态估计融合算法[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 95-103.
[12] 胡强, 刘倩, 周杭霞. 基于改进Stacking策略的钓鱼网站检测研究[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 132-140.
[13] 段美玲, 潘巨龙. 基于双向LSTM神经网络可穿戴跌倒检测研究[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 141-150.
[14] 吴军, 欧阳艾嘉, 张琳. 基于多头注意力机制的磷酸化位点预测模型[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 161-171.
[15] 闫龙川, 李妍, 宋浒, 邹昊东, 王丽君. 基于Prophet-DeepAR模型的Web流量预测[J]. 广西师范大学学报（自然科学版）, 2022, 40(3): 172-184.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
版权所有 © 广西师范大学学报（自然科学版）编辑部
地址：广西桂林市三里店育才路15号　邮编：541004
电话：0773-5857325　E-mail: gxsdzkb@mailbox.gxnu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发