[1]郭大伟,李景浩,陆军.三维点云配准的多尺度深度学习方法[J].智能系统学报,2024,19(4):817-826.[doi:10.11992/tis.202304007]
GUO Dawei,LI Jinghao,LU Jun.Multi-scale deep learning method for 3D point cloud registration[J].CAAI Transactions on Intelligent Systems,2024,19(4):817-826.[doi:10.11992/tis.202304007]
点击复制
GUO Dawei,LI Jinghao,LU Jun.Multi-scale deep learning method for 3D point cloud registration[J].CAAI Transactions on Intelligent Systems,2024,19(4):817-826.[doi:10.11992/tis.202304007]
三维点云配准的多尺度深度学习方法
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
19
期数:
2024年第4期
页码:
817-826
栏目:
学术论文—机器学习
出版日期:
2024-07-05
- Title:
- Multi-scale deep learning method for 3D point cloud registration
- Keywords:
- point cloud registration; 3D vision; feature extraction; feature fusion; deep learning; point of interest aggregation; multi-scale features; Transformer network
- 分类号:
- TP391.4
- 摘要:
- 针对近几年三维视觉领域基于深度学习点云配准算法鲁棒性差和精度较差等问题，设计了一种基于深度学习的三维点云配准方法。首先抽取具有明显几何特征的点作为兴趣点，通过区域生长算法对兴趣点进行聚合，并基于多尺度分析方法进行特征提取以及特征融合。为进一步提取特征数据中包含的深层局部信息，使用多层感知机（multilayer perceptron，MLP）网络进行二次特征提取，并加入修改过的Transformer网络补充特征。设计了匹配矩阵生成及优化算法，并通过奇异值分解（singular value decomposition，SVD）计算得到变换矩阵。通过在ModelNet40数据集上进行比较实验，证明本文的配准算法远优于传统配准算法，并在配准精度和鲁棒性方面优于近几年流行的深度配准网络DCPNet和RPMNet。本文分析结果可为提高点云配准鲁棒性以及精度提供参考。
- Abstract:
- In order to solve the problems of poor robustness and poor accuracy of the point cloud registration algorithm based on deep learning in the field of 3D vision in recent years, a 3D point cloud registration method based on deep learning is designed. Firstly, the points with obvious geometric features are extracted as points of interest, and the points of interest are aggregated by the regional growth algorithm, and feature extraction and feature fusion are carried out based on the multi-scale analysis method. In order to further extract the deep local information contained in the feature data, the multilayer perceptron(MLP) network is used for secondary feature extraction, and the modified Transformer network is added to supplement the features. The matching matrix generation and optimization algorithm is designed, and the transformation matrix is obtained by singular value decomposition(SVD) calculation. Through comparative experiments on the ModelNet40 dataset, it is proved that the registration algorithm in this paper is far superior to the traditional registration algorithm, and is better than the popular deep registration networks DCPNet and RPMNet in recent years in terms of registration accuracy and robustness. The analysis results in this paper provide a reference for improving the robustness and accuracy of point cloud registration.
- 参考文献/References:
-
[1] REN Tianyu, WU Ruicheng. An acceleration algorithmof 3D point cloud registration based on iterative closet point[C]//2020 Asia-Pacific Conference on Image Processing, Electronics and Computers. Dalian: IEEE, 2020: 271-276.
[2] FENG Shanshan, LIN Yun, WANG Yanping, et al. 3D point cloud reconstruction using inversely mapping and voting from single pass CSAR images[J]. Remote sensing, 2021, 13(17): 3534.
[3] 李姗姗, 张娜娜, 张媛媛, 等. 基于MSD-Vnet的三维医学图像配准[J]. 电视技术, 2021, 45(1): 51–56
LI Shanshan, ZHANG Nana, ZHANG Yuanyuan, et al. Three-dimensional medical image registration based on MSD-Vnet[J]. Television technology, 2021, 45(1): 51–56
[4] 鹿道玺. 基于多源信息融合的SLAM算法研究[J]. 无线互联科技, 2022, 19(24): 141–144
LU Daoxi. Research on SLAM algorithm based on multi-source information fusion[J]. Wireless internet technology, 2022, 19(24): 141–144
[5] SAYED H, AHMED E, AHMED S, et al. A computer-aided diagnostic system for diabetic retinopathy based on local and global extracted features[J]. Applied sciences, 2022, 12(16): 8326–8326.
[6] QI C R, SU Hao, MO Kaichun, et al. Pointnet: deep learning on point sets for 3d classification and segmentation[C]//Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 52-660.
[7] ZENG Andy, SONG Shuran, NIEBNER M, et al. 3DMatch: learning local geometric descriptors from RGB-dreconstructions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 1802-1811.
[8] ELBAZ G, AVRAHAM T, FISCHER A. 3D point cloud registration for localization using a deep neuralnetwork auto-encoder[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Honolulu: IEEE, 2017: 4631-4640.
[9] LI J, LEE G H. Usip: unsupervised stable interest point detection from 3d point clouds[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 361-370.
[10] LU Fan, CHEN Guang, LIU Yinlong, et al. Rskdd-net: random sample-based keypoint detector and descriptor[EB/OL]. (2020-10-23)[2023-04-01]. https://doi.org/10.48550/arXiv.2010.12394.
[11] YEW Z J, LEE G H. 3D featnet: weakly supervised local 3d features for point cloud registration[C]//Proceedings of the European Conference on Computer Vision. Berlin: Springer, 2018: 607-623.
[12] SCHROFF F, KALENICHENKO D, PHILBIN J. Facenet: a unified embedding for face recognition and clustering[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 815-823.
[13] ELIZEU M O J, DANIEL R S, GIOVANA A R M. A new variant of the ICP algorithm for pairwise 3D point cloud registration[J]. American academic scientific research journal for engineering, technology, and sciences, 2022, 85(1): 71–88.
[14] MYRONENKO A, SONG X. Point set registration: coherent point drift[J]. IEEE transactions on pattern analysis and machine intelligence, 2010, 32(12): 2262–2275.
[15] GAO W, TEDRAKE R. Filterreg: robust and efficient probabilistic point-set registration using gaussian filter and twist parameterization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 11095-11104.
[16] JIANG Huijun, LI Quefeng, LIN J T, et al. Classification of disease recurrence using transition likelihoods with expectation-maximization algorithm[J]. Statistics in medicine, 2022, 41(23): 4697–4715.
[17] LI Ding, CHEN Feng. Deep mapping: unsupervised map estimation from multiple point clouds[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 8650-8659.
[18] AOKI Y, GOFORTH H, SRIVATSAN R A, et al. Pointnetlk: robust & efficient point cloud registrationusing pointnet[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 7163-7172.
[19] 林桂潮, 张青, 邹湘军. 基于改进Lucas-Kanade的亚像素级零件图像配准[J]. 计算机应用研究, 2017, 34(5): 1577–1580, 1593
LIN Guichao, ZHANG Qing, ZOU Xiangjun. Subpixel level parts image registration based on improved lucas-kanade[J]. Application research of computers, 2017, 34(5): 1577–1580, 1593
[20] SARODE V, LI X, GOFORTH H, et al. PCRNet: point cloud registration network using pointNet encoding[EB/OL]. (2019-08-21)[2023-04-01]. https://doi.org/10.48550/arXiv.1908.07906.
[21] WANG Y, SOLOMON J M. Deep closest point: learning representations for point cloud registration[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 3523-3532.
[22] GU Fengwei, LU Jun, CAI Chengtao. RPformer: a robust parallel transformer for visual tracking in complex scenes[J]. IEEE transactions on instrumentation and measurement, 2022, 71: 1–14.
[23] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[EB/OL]. (2017-06-12)[2023-04-01]. https://doi.org/10.48550/arXiv.1706.03762.
[24] 何翠萍. 应用区域生长的无人船红外图像精确分割方法[J]. 信息与电脑(理论版), 2022, 34(24): 81–83
HE Cuiping. Accurate segmentation method of infrared image of unmanned ship growing in application area[J]. Information and computer(theoretical edition), 2022, 34(24): 81–83
[25] YU Zhong, Intrinsic shape signatures: a shape descriptor for 3D object recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Kyoto: IEEE, 2009: 689-696.
- 相似文献/References:
-
[1]张毅,陈起,罗元.室内环境下移动机器人三维视觉SLAM[J].智能系统学报,2015,10(4):615.[doi:10.3969/j.issn.1673-4785.201504003]
ZHANG Yi,CHEN Qi,LUO Yuan.Three dimensional visual SLAM for mobile robots in indoor environments[J].CAAI Transactions on Intelligent Systems,2015,10():615.[doi:10.3969/j.issn.1673-4785.201504003]
备注/Memo
收稿日期:2023-04-04。
基金项目:黑龙江省自然科学基金项目（F201123）.
作者简介:郭大伟，硕士研究生，主要研究方向为人脸识别、机器视觉、图像处理。E-mail：2781005851@qq.com;陆军，教授，博士生导师，主要研究方向为计算机视觉、智能控制技术、图像处理、高性能船舶控制和机械臂控制。获国防科学技术进步奖一等奖3项，省部级二等奖1项，省部级三等奖3项，授权发明专利10项。发表学术论文70余篇。E-mail：lujun0260@sina. com
通讯作者:陆军. E-mail：lujun0260@sina.com
更新日期/Last Update:
1900-01-01