面向点云配准和地点识别的多头旋转注意力网络

点云配准和地点识别是移动机器人和自动驾驶车辆实现自主定位的关键技术.目前鲜有方法能够在实现高效地点识别的同时输出准确的6自由度位姿.本文提出了一种新颖的多头网络,该网络首先在统一的主干网络中提取稀疏而特征显著的点,随后分别在稠密点匹配头中解决点云配准问题,在全局描述头中解决地点识别问题.其中,在主干网络中创新应用3D-RoFormer机制,以一种低计算和存储复杂度的方式显式地编码特征点之间的相对位姿信息,从而学习到更显著和鲁棒的点特征,有效提高了网络的特征表达能力.在稠密点匹配头中,首先,构建稀疏点可靠的匹配关系,并据此由粗至精地确定稠密点的匹配关系,进而优化位姿估计.在全局描述头中,网络将稀...

Full description

Saved in:

Bibliographic Details
Published in	控制理论与应用 Vol. 40; no. 12; pp. 2187 - 2197
Main Authors	施成浩, 陈谢沅澧, 郭瑞斌, 肖军浩, 戴斌, 卢惠民
Format	Journal Article
Language	Chinese
Published	国防科技大学智能科学学院,湖南长沙 410073%国防科技创新研究院无人系统技术研究中心,北京 100091 01.12.2023
Subjects	闭环 deep learning autonomous vehicles 地点识别 place recognition loop closing 自动驾驶汽车深度学习 3D registration 三维点云配准
Online Access	Get full text
ISSN	1000-8152
DOI	10.7641/CTA.2023.30286

Cover

Abstract	点云配准和地点识别是移动机器人和自动驾驶车辆实现自主定位的关键技术.目前鲜有方法能够在实现高效地点识别的同时输出准确的6自由度位姿.本文提出了一种新颖的多头网络,该网络首先在统一的主干网络中提取稀疏而特征显著的点,随后分别在稠密点匹配头中解决点云配准问题,在全局描述头中解决地点识别问题.其中,在主干网络中创新应用3D-RoFormer机制,以一种低计算和存储复杂度的方式显式地编码特征点之间的相对位姿信息,从而学习到更显著和鲁棒的点特征,有效提高了网络的特征表达能力.在稠密点匹配头中,首先,构建稀疏点可靠的匹配关系,并据此由粗至精地确定稠密点的匹配关系,进而优化位姿估计.在全局描述头中,网络将稀疏的特征点及其特征向量进行压缩编码,获得对相关点云的全局描述子,实现高效的地点识别.为了验证算法的有效性并评估其性能,本文针对不同环境、不同传感器获得的数据集开展了实验研究.实验结果表明,本文方法在所有测试数据集中都具有很好的泛化能力,并较当前先进方法都有更优或相当的表现,降低连续点云配准误差约27％,降低闭环点云配准误差约37％.
AbstractList	点云配准和地点识别是移动机器人和自动驾驶车辆实现自主定位的关键技术.目前鲜有方法能够在实现高效地点识别的同时输出准确的6自由度位姿.本文提出了一种新颖的多头网络,该网络首先在统一的主干网络中提取稀疏而特征显著的点,随后分别在稠密点匹配头中解决点云配准问题,在全局描述头中解决地点识别问题.其中,在主干网络中创新应用3D-RoFormer机制,以一种低计算和存储复杂度的方式显式地编码特征点之间的相对位姿信息,从而学习到更显著和鲁棒的点特征,有效提高了网络的特征表达能力.在稠密点匹配头中,首先,构建稀疏点可靠的匹配关系,并据此由粗至精地确定稠密点的匹配关系,进而优化位姿估计.在全局描述头中,网络将稀疏的特征点及其特征向量进行压缩编码,获得对相关点云的全局描述子,实现高效的地点识别.为了验证算法的有效性并评估其性能,本文针对不同环境、不同传感器获得的数据集开展了实验研究.实验结果表明,本文方法在所有测试数据集中都具有很好的泛化能力,并较当前先进方法都有更优或相当的表现,降低连续点云配准误差约27％,降低闭环点云配准误差约37％.
Abstract_FL	Point cloud registration and place recognition are critical tasks for localization in robotics and autonomous driving.There are few methods that can achieve efficient place recognition while providing accurate 6-degree-of-freedom pose.In this paper,we propose a novel multi-head network that simultaneously addresses both of these tasks.The network first extracts discriminative sparse points from the point cloud using a backbone network,and then solves the point cloud registration task in a dense point matching head and the place recognition task in a global descriptor head.In the backbone,we apply a novel 3D-RoFormer mechanism that explicitly encodes the relative pose information of points efficiently,result-ing in more discriminative and robust point features and significantly improving network performance.In the dense point matching head,the network establishes reliable correspondences between sparse points and progressively finds coarse-to-fine dense point correspondences to improve final pose estimation.In the global descriptor head,the network compresses the sparse point features into a global descriptor to describe the features of the current point cloud and achieves place recognition.We extensively evaluate our method on multiple datasets collected by different sensors in various environ-ments.Experimental results show that our method depicts strong generalization ability on all the datasets,outperforming or performing comparably to the state-of-the-art methods,among which the continuous point cloud registration error is reduced by about 27％,and the closed-loop point cloud registration error is reduced by about 37％.
Author	陈谢沅澧肖军浩戴斌施成浩郭瑞斌卢惠民
AuthorAffiliation	国防科技大学智能科学学院,湖南长沙 410073%国防科技创新研究院无人系统技术研究中心,北京 100091
AuthorAffiliation_xml	– name: 国防科技大学智能科学学院,湖南长沙 410073%国防科技创新研究院无人系统技术研究中心,北京 100091
Author_FL	XIAO Jun-hao SHI Cheng-hao LU Hui-min DAI Bin CHEN Xie-yuanli GUO Rui-bin
Author_FL_xml	– sequence: 1 fullname: SHI Cheng-hao – sequence: 2 fullname: CHEN Xie-yuanli – sequence: 3 fullname: GUO Rui-bin – sequence: 4 fullname: XIAO Jun-hao – sequence: 5 fullname: DAI Bin – sequence: 6 fullname: LU Hui-min
Author_xml	– sequence: 1 fullname: 施成浩 – sequence: 2 fullname: 陈谢沅澧 – sequence: 3 fullname: 郭瑞斌 – sequence: 4 fullname: 肖军浩 – sequence: 5 fullname: 戴斌 – sequence: 6 fullname: 卢惠民
BookMark	eNotjz1Lw1AYRu9QwVq7-htcEt_33nxOUoJfUHCpc7nJzRU1pGAQiaPUKlbRuAi1LtJBUFDQRaX_xpu0_8KITg-c4RyeOVKJO3FIyAKCblsGLnmthk6BMp0BdawKqSIAaA6adJbUk2THB0AEaptYJcvT-wd1nRXHH9-f2fTkUp321M2FGr6WZPLSU2dPxaCrRgM1es9v-5Pxc_72mHev1PldMc6Kr-E8mZE8SsL6_9bI1upKy1vXmptrG16jqSUIrIxzdEWAIpQWEwZ1TGH4DBCQ-07gS-QoAorMFOgabmiElFGLSsZtCcINbJfVyOKf95DHksfb7d3OwX5cFtt7R1GUpunvYaSlkv0Axfxh6Q
ContentType	Journal Article
Copyright	Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
Copyright_xml	– notice: Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
DBID	2B. 4A8 92I 93N PSX TCJ
DOI	10.7641/CTA.2023.30286
DatabaseName	Wanfang Data Journals - Hong Kong WANFANG Data Centre Wanfang Data Journals 万方数据期刊 - 香港版 China Online Journals (COJ) China Online Journals (COJ)
DatabaseTitleList
DeliveryMethod	fulltext_linktorsrc
Discipline	Engineering
DocumentTitle_FL	A novel multiplex rotational attention-based network for point cloud registration and place recognition
EndPage	2197
ExternalDocumentID	kzllyyy202312010
GroupedDBID	-01 -0Y 2B. 4A8 5XA 5XJ 92H 92I 93N ABJNI ACGFS ALMA_UNASSIGNED_HOLDINGS CCEZO CCVFK CUBFJ CW9 PSX TCJ TGT U1G U5S UY8
ID	FETCH-LOGICAL-s1030-8a19dc1def63d4285d4b30101ab8cbf1a1dc2135d1949e4e23262f3a7f0d9c793
ISSN	1000-8152
IngestDate	Thu May 29 04:08:30 EDT 2025
IsPeerReviewed	true
IsScholarly	true
Issue	12
Keywords	闭环 deep learning autonomous vehicles 地点识别 place recognition loop closing 自动驾驶汽车深度学习 3D registration 三维点云配准
Language	Chinese
LinkModel	OpenURL
MergedId	FETCHMERGED-LOGICAL-s1030-8a19dc1def63d4285d4b30101ab8cbf1a1dc2135d1949e4e23262f3a7f0d9c793
PageCount	11
ParticipantIDs	wanfang_journals_kzllyyy202312010
PublicationCentury	2000
PublicationDate	2023-12-01
PublicationDateYYYYMMDD	2023-12-01
PublicationDate_xml	– month: 12 year: 2023 text: 2023-12-01 day: 01
PublicationDecade	2020
PublicationTitle	控制理论与应用
PublicationTitle_FL	Control Theory & Applications
PublicationYear	2023
Publisher	国防科技大学智能科学学院,湖南长沙 410073%国防科技创新研究院无人系统技术研究中心,北京 100091
Publisher_xml	– name: 国防科技大学智能科学学院,湖南长沙 410073%国防科技创新研究院无人系统技术研究中心,北京 100091
SSID	ssib001102751 ssib002258297 ssib023646306 ssib057620041 ssib051372463 ssj0042201 ssib023167526
Score	2.3988905
Snippet	...
SourceID	wanfang
SourceType	Aggregation Database
StartPage	2187
Title	面向点云配准和地点识别的多头旋转注意力网络
URI	https://d.wanfangdata.com.cn/periodical/kzllyyy202312010
Volume	40
hasFullText	1
inHoldings	1
isFullTextHit
isPrint
link	http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnV3Na9RAFA-lvehB_MRvKjinsjWZTCYzJ5nZzVLEerGF3iTZZBUsK9j20B6lreIHWi9CrRfpQVBQ0ItK_xvTj__C916S3bT0UL0ss5M37_3ex8682cyH41yPpS87bpI0Aq3ihlA6bSg_CRteN-2EWcDjNMMNzpN35MS0uDUTzAwNr9RWLS3MJ-OdpUP3lfyPV6EO_Iq7ZP_Bs32mUAFl8C98gofh80g-ZpFmusUMZ1HAtMu0x6KQKc6sZpFg1lCNZipgqoU0MPNXkog5U00qNJl1a60UM-2SRkHZ4iNtmBJYYwSWi4KFGsk0NLTYygKMJtZYnxmFBWii2sQHYBAfoCkQWgty62kx0UfMhKVcK0muS0gARoS6oEYKyYAGVRNEI0BcFTWESKIc5KeQASICvHpAAibT-BBBu2Q7IOFoIywgiDqtAnVaJMhjOioFqOaARKHldGEwSYrWJZb_p3D_wNoUMr0lpIBHIQCQYUgMYje4AqUwdGEUAGEkiddkC4XIbKvWqk_TL5CmoCAvHKMrmC10Gz4FO7ZL9bUeE7iUxWc8OCo4tLCtbE5BZNzSLUaTDysERaTA0yIqUYEQI8XaKhzaNc4SoxLC8BCGFAKGYhmggwk4RTHEMqpEzE1zDN8mFbezleMtHmygvGDfgFyc31V1PHzf8FpmZ1k12oaHpQGhFJgGNKfMOHp43Ick-sB565TBPVyanV1cXEQaj9NGzREehrjaY8S0Jm_fHcwrPHy7X8tzeYBb0qvvHI-TCAbzGLyUQfqD5QuB54e8du0EzPGxG-uvRROc0z3tfWsUp8uiGjf2K0FbC3vduHe_lgVPnXROlNPXUVP0RaecoaUHp53jtUNNzzg39z58zN-s7Tz5-efX2t7Kq_zpav72Zb7xDWp2v67mzz7vrC_nm-v55o_tdy92t75sf_-0vfw6f_5-Z2tt5_fGWWe6HU01JxrlNS2NObyjsKFiT6cdDzp16aeCqyAViY9HV8aJ6iRdL_bSDvf8IPW00JnIYA4nedePw66b6g7kB-ec4d6jXnbeGZVKwnzIFzBKuCKO_USrJPBSkeECmUzGF5xrpfr3ym547t5BJ148As0l59jgp3_ZGZ5_vJBdgcnFfHK1dP1fjNfO8w
linkProvider	EBSCOhost
openUrl	ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=%E9%9D%A2%E5%90%91%E7%82%B9%E4%BA%91%E9%85%8D%E5%87%86%E5%92%8C%E5%9C%B0%E7%82%B9%E8%AF%86%E5%88%AB%E7%9A%84%E5%A4%9A%E5%A4%B4%E6%97%8B%E8%BD%AC%E6%B3%A8%E6%84%8F%E5%8A%9B%E7%BD%91%E7%BB%9C&rft.jtitle=%E6%8E%A7%E5%88%B6%E7%90%86%E8%AE%BA%E4%B8%8E%E5%BA%94%E7%94%A8&rft.au=%E6%96%BD%E6%88%90%E6%B5%A9&rft.au=%E9%99%88%E8%B0%A2%E6%B2%85%E6%BE%A7&rft.au=%E9%83%AD%E7%91%9E%E6%96%8C&rft.au=%E8%82%96%E5%86%9B%E6%B5%A9&rft.date=2023-12-01&rft.pub=%E5%9B%BD%E9%98%B2%E7%A7%91%E6%8A%80%E5%A4%A7%E5%AD%A6%E6%99%BA%E8%83%BD%E7%A7%91%E5%AD%A6%E5%AD%A6%E9%99%A2%2C%E6%B9%96%E5%8D%97%E9%95%BF%E6%B2%99+410073%25%E5%9B%BD%E9%98%B2%E7%A7%91%E6%8A%80%E5%88%9B%E6%96%B0%E7%A0%94%E7%A9%B6%E9%99%A2%E6%97%A0%E4%BA%BA%E7%B3%BB%E7%BB%9F%E6%8A%80%E6%9C%AF%E7%A0%94%E7%A9%B6%E4%B8%AD%E5%BF%83%2C%E5%8C%97%E4%BA%AC+100091&rft.issn=1000-8152&rft.volume=40&rft.issue=12&rft.spage=2187&rft.epage=2197&rft_id=info:doi/10.7641%2FCTA.2023.30286&rft.externalDocID=kzllyyy202312010
thumbnail_s	http://utb.summon.serialssolutions.com/2.0.0/image/custom?url=http%3A%2F%2Fwww.wanfangdata.com.cn%2Fimages%2FPeriodicalImages%2Fkzllyyy%2Fkzllyyy.jpg