en
×

分享给微信好友或者朋友圈

使用微信“扫一扫”功能。
作者简介:

尚光涛,男,硕士生,研究方向为移动机器人、SLAM.20201222014@nuist.edu.cn

通讯作者:

陈炜峰,男,博士,教授,研究方向为机器人、机电控制.002021@nuist.edu.cn

中图分类号:TP242;TP391.41

文献标识码:A

DOI:10.13878/j.cnki.jnuist.20220420001

参考文献 1
任伟建,高强,康朝海,等.移动机器人同步定位与建图技术综述[J].计算机测量与控制,2022,30(2):1-10,37.REN Weijian,GAO Qiang,KANG Chaohai,et al.Overview of simultaneous localization and mapping technology of mobile robots[J].Computer Measurement & Control,2022,30(2):1-10,37
参考文献 2
赵乐文,任嘉倩,丁杨.基于GNSS的空间环境参数反演平台及精度评估[J].南京信息工程大学学报(自然科学版),2021,13(2):204-210.ZHAO Lewen,REN Jiaqian,DING Yang.Platform for GNSS real-time space environment parameter inversion and its accuracy evaluation[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2021,13(2):204-210
参考文献 3
尹姝,陈元橼,仇翔.基于RFID和自适应卡尔曼滤波的室内移动目标定位方法[J].南京信息工程大学学报(自然科学版),2018,10(6):749-753.YIN Shu,CHEN Yuanyuan,QIU Xiang.Indoor moving-target localization using RFID and adaptive Kalman filter[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2018,10(6):749-753
参考文献 4
周韦,孙宪坤,吴飞.基于SLAM/UWB的室内融合定位算法研究[J].全球定位系统,2022,47(1):36-42,85.ZHOU Wei,SUN Xiankun,WU Fei.Research on indoor fusion positioning algorithm based on SLAM/UWB[J].GNSS World of China,2022,47(1):36-42,85
参考文献 5
Bresson G,Alsayed Z,Yu L,et al.Simultaneous localization and mapping:a survey of current trends in autonomous driving[J].IEEE Transactions on Intelligent Vehicles,2017,2(3):194-220
参考文献 6
李晓飞,宋亚男,徐荣华,等.基于双目视觉的船舶跟踪与定位[J].南京信息工程大学学报(自然科学版),2015,7(1):46-52.LI Xiaofei,SONG Yanan,XU Ronghua,et al.Tracking and positioning of ship based on binocular vision[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2015,7(1):46-52
参考文献 7
刘明芹,张晓光,徐桂云,等.单机器人SLAM技术的发展及相关主流技术综述[J].计算机工程与应用,2020,56(18):25-35.LIU Mingqin,ZHANG Xiaoguang,XU Guiyun,et al.Review of development of single robot SLAM technology and related mainstream technology[J].Computer Engineering and Applications,2020,56(18):25-35
参考文献 8
权美香,朴松昊,李国.视觉SLAM综述[J].智能系统学报,2016(6):768-776.QUAN Meixiang,PIAO Songhao,LI Guo.An overview of visual SLAM[J].CAAI Transactions on Intelligent Systems,2016(6):768-776
参考文献 9
胡凯,吴佳胜,郑翡,等.视觉里程计研究综述[J].南京信息工程大学学报(自然科学版),2021,13(3):269-280.HU Kai,WU Jiasheng,ZHENG Fei,et al.A survey of visual odometry[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2021,13(3):269-280
参考文献 10
刘瑞军,王向上,张晨,等.基于深度学习的视觉SLAM综述[J].系统仿真学报,2020,32(7):1244-1256.LIU Ruijun,WANG Xiangshang,ZHANG Chen,et al.A survey on visual SLAM based on deep learning[J].Journal of System Simulation,2020,32(7):1244-1256
参考文献 11
李少朋,张涛.深度学习在视觉SLAM中应用综述[J].空间控制技术与应用,2019,45(2):1-10.LI Shaopeng,ZHANG Tao.A survey of deep learning application in visual SLAM[J].Aerospace Control and Application,2019,45(2):1-10
参考文献 12
Davison A J,Reid I D,Molton N D,et al.MonoSLAM:real-time single camera SLAM[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2007,29(6):1052-1067
参考文献 13
Newcombe R A,Lovegrove S J,Davison A J.DTAM:dense tracking and mapping in real-time[C]//2011 International Conference on Computer Vision.November 6-13,2011,Barcelona,Spain.IEEE,2011:2320-2327
参考文献 14
Mur-Artal R,Montiel J M M,Tardos J D.ORB-SLAM:a versatile and accurate monocular SLAM system[J].IEEE Transactions on Robotics,2015,31(5):1147-1163
参考文献 15
Galvez-Lopez D,Tardos J D.Bags of binary words for fast place recognition in image sequences[J].IEEE Transactions on Robotics,2012,28(5):1188-1197
参考文献 16
Mur-Artal R,Tardos J D.ORB-SLAM2:an open-source SLAM system for monocular,stereo,and RGB-D cameras[J].IEEE Transactions on Robotics,2017,33(5):1255-1262
参考文献 17
Campos C,Elvira R,Rodriguez J J G,et al.ORB-SLAM3:an accurate open-source library for visual,visual-inertial,and multimap SLAM[J].IEEE Transactions on Robotics,2021,37(6):1874-1890
参考文献 18
Engel J,Koltun V,Cremers D.Direct sparse odometry[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(3):611-625
参考文献 19
Qin T,Li P L,Shen S J.VINS-mono:a robust and versatile monocular visual-inertial state estimator[J].IEEE Transactions on Robotics,2018,34(4):1004-1020
参考文献 20
邓晨,李宏伟,张斌,等.基于深度学习的语义SLAM关键帧图像处理[J].测绘学报,2021,50(11):1605-1616.DENG Chen,LI Hongwei,ZHANG Bin,et al.Research on key frame image processing of semantic SLAM based on deep learning[J].Acta Geodaetica et Cartographica Sinica,2021,50(11):1605-1616
参考文献 21
Klein G,Murray D.Parallel tracking and mapping for small AR workspaces[C]//2007 6th IEEE and ACM International Symposium on Mixed and Augmented Reality.November 13-16,2017,Nara,Japan.IEEE,2007:225-234
参考文献 22
Gomez-Ojeda R,Briales J,Gonzalez-Jimenez J.PL-SVO:semi-direct monocular visual odometry by combining points and line segments[C]//2016 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).October 9-14,2016,Daejeon,Korea(South).IEEE,2016:4211-4216
参考文献 23
Pumarola A,Vakhitov A,Agudo A,et al.PL-SLAM:real-time monocular visual SLAM with points and lines[C]//2017 IEEE International Conference on Robotics and Automation.May 29-June 3,2017,Singapore.IEEE,2017:4503-4508
参考文献 24
Forster C,Pizzoli M,Scaramuzza D.SVO:fast semi-direct monocular visual odometry[C]//2014 IEEE International Conference on Robotics and Automation.May 31-June 7,2014,Hong Kong,China.IEEE,2014:15-22
参考文献 25
Engel J,Schps T,Cremers D.LSD-SLAM:large-scale direct monocular SLAM[C]//European Conference on Computer Vision.Springer,Cham,2014:834-849
参考文献 26
Mourikis A I,Roumeliotis S I.A multi-state constraint Kalman filter for vision-aided inertial navigation[C]//Proceedings 2007 IEEE International Conference on Robotics and Automation.April 10-14,2007,Rome,Italy.IEEE,2007:3565-3572
参考文献 27
Leutenegger S,Lynen S,Bosse M,et al.Keyframe-based visual-inertial odometry using nonlinear optimization[J].The International Journal of Robotics Research,2015,34(3):314-334
参考文献 28
Bloesch M,Omari S,Hutter M,et al.Robust visual inertial odometry using a direct EKF-based approach[C]//2015 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 28-October 2,2015,Hamburg,Germany.IEEE,2015:298-304
参考文献 29
Xu D,Vedaldi A,Henriques J F.Moving SLAM:fully unsupervised deep learning in non-rigid scenes[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 27-October 1,2021,Prague,Czech Republic.IEEE,2021:4611-4617
参考文献 30
张彦雯,胡凯,王鹏盛.三维重建算法研究综述[J].南京信息工程大学学报(自然科学版),2020,12(5):591-602.ZHANG Yanwen,HU Kai,WANG Pengsheng.Review of 3D reconstruction algorithms[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2020,12(5):591-602
参考文献 31
Li J L,Li Z J,Feng Y,et al.Development of a human-robot hybrid intelligent system based on brain teleoperation and deep learning SLAM[J].IEEE Transactions on Automation Science and Engineering,2019,16(4):1664-1674
参考文献 32
Mumuni A,Mumuni F.CNN architectures for geometric transformation-invariant feature representation in computer vision:a review[J].SN Computer Science,2021,2(5):1-23
参考文献 33
Ma R B,Wang R,Zhang Y B,et al.RNNSLAM:reconstructing the 3D colon to visualize missing regions during a colonoscopy[J].Medical Image Analysis,2021,72:102100
参考文献 34
Tateno K,Tombari F,Laina I,et al.CNN-SLAM:real-time dense monocular SLAM with learned depth prediction[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6565-6574
参考文献 35
Wang S,Clark R,Wen H K,et al.DeepVO:towards end-to-end visual odometry with deep recurrent convolutional neural networks[C]//2017 IEEE International Conference on Robotics and Automation.May 29-June 3,2017,Singapore.IEEE,2017:2043-2050
参考文献 36
Bloesch M,Czarnowski J,Clark R,et al.CodeSLAM:learning a compact,optimisable representation for dense visual SLAM[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.June 18-23,2018,Salt Lake City,UT,USA.IEEE,2018:2560-2568
参考文献 37
Yang N,Wang R,Stuckler J,et al.Deep virtual stereo odometry:leveraging deep depth prediction for monocular direct sparse odometry[C]//European Conference on Computer Vision(ECCV).September 8-12,2018,Munich,Germany.2018.DOI:10.48550/arXiv.1807.02570
参考文献 38
Li R H,Wang S,Long Z Q,et al.UnDeepVO:monocular visual odometry through unsupervised deep learning[C]//2018 IEEE International Conference on Robotics and Automation.May 21-25,2018,Brisbane,QLD,Australia.IEEE,2018:7286-7291
参考文献 39
Loo S Y,Amiri A J,Mashohor S,et al.CNN-SVO:improving the mapping in semi-direct visual odometry using single-image depth prediction[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5218-5223
参考文献 40
Almalioglu Y,Saputra M R U,de Gusmão P P B,et al.GANVO:unsupervised deep monocular visual odometry and depth estimation with generative adversarial networks[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5474-5480
参考文献 41
Li Y,Ushiku Y,Harada T.Pose graph optimization for unsupervised monocular visual odometry[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5439-5445
参考文献 42
Yang N,von Stumberg L,Wang R,et al.D3VO:deep depth,deep pose and deep uncertainty for monocular visual odometry[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).June 13-19,2020,Seattle,WA,USA.IEEE,2020:1278-1289
参考文献 43
Chancán M,Milford M.DeepSeqSLAM:a trainable CNN+RNN for joint global description and sequence-based place recognition[J].arXiv e-print,2020,arXiv:2011.08518
参考文献 44
Li R H,Wang S,Gu D B.DeepSLAM:a robust monocular SLAM system with unsupervised deep learning[J].IEEE Transactions on Industrial Electronics,2021,68(4):3577-3587
参考文献 45
Bruno H M S,Colombini E L.LIFT-SLAM:a deep-learning feature-based monocular visual SLAM method[J].Neurocomputing,2021,455:97-110
参考文献 46
Zhang S M,Lu S Y,He R,et al.Stereo visual odometry pose correction through unsupervised deep learning[J].Sensors(Basel,Switzerland),2021,21(14):4735
参考文献 47
Clark R,Wang S,Wen H,et al.VINet:visual-inertial odometry as a sequence-to-sequence learning problem[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence.February 4-9,2017,San Francisco,CA,USA.2017:3995-4001
参考文献 48
Shamwell E J,Lindgren K,Leung S,et al.Unsuperviseddeep visual-inertial odometry with online error correction for RGB-D imagery[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(10):2478-2493
参考文献 49
Han L M,Lin Y M,Du G G,et al.DeepVIO:self-supervised deep learning of monocular visual inertial odometry using 3D geometric constraints[C]//2019 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).November 3-8,2019,Macao,China.IEEE,2019:6906-6913
参考文献 50
Chen C H,Rosa S,Miao Y S,et al.Selective sensor fusion for neural visual-inertial odometry[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).June 15-20,2019,Long Beach,CA,USA.IEEE,2019:10534-10543
参考文献 51
Kim Y,Yoon S,Kim S,et al.Unsupervised balanced covariance learning for visual-inertial sensor fusion[J].IEEE Robotics and Automation Letters,2021,6(2):819-826
参考文献 52
Gurturk M,Yusefi A,Aslan M F,et al.The YTU dataset and recurrent neural network based visual-inertial odometry[J].Measurement,2021,184:109878
参考文献 53
傅杰,徐常胜.关于单目标跟踪方法的研究综述[J].南京信息工程大学学报(自然科学版),2019,11(6):638-650.FU Jie,XU Changsheng.A survey of single object tracking methods[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2019,11(6):638-650
参考文献 54
Steenbeek A,Nex F.CNN-based dense monocular visual SLAM for real-time UAV exploration in emergency conditions[J].Drones,2022,6(3):79
参考文献 55
唐灿,唐亮贵,刘波.图像特征检测与匹配方法研究综述[J].南京信息工程大学学报(自然科学版),2020,12(3):261-273.TANG Can,TANG Lianggui,LIU Bo.A survey of image feature detection and matching methods[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2020,12(3):261-273
参考文献 56
Li L,Kong X,Zhao XR,et al.Semantic scan context:a novel semantic-based loop-closure method for LiDAR SLAM[J].Autonomous Robots,2022,46(4):535-551
参考文献 57
Sakkari M,Hamdi M,Elmannai H,et al.Feature extraction-based deep self-organizing map[J].Circuits,Systems,and Signal Processing,2022,41(5):2802-2824
参考文献 58
Zhu R,Yang M K,Liu W,et al.DeepAVO:efficient pose refining with feature distilling for deep visual odometry[J].Neurocomputing,2022,467:22-35
参考文献 59
Kim J J Y,Urschler M,Riddle P J,et al.SymbioLCD:ensemble-based loop closure detection using CNN-extracted objects and visual bag-of-words[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 27-October 1,2021,Prague,Czech Republic.IEEE,2021:5425
参考文献 60
Ai Y B,Rui T,Lu M,et al.DDL-SLAM:a robust RGB-D SLAM in dynamic environments combined with deep learning[J].IEEE Access,8:162335-162342
参考文献 61
Javed Z,Kim G W.PanoVILD:a challenging panoramic vision,inertial and LiDAR dataset for simultaneous localization and mapping[J].The Journal of Supercomputing,2022,78(6):8247-8267
参考文献 62
Duan R,Feng Y R,Wen C Y.Deep pose graph-matching-based loop closure detection for semantic visual SLAM[J].Sustainability,2022,14(19):11864
参考文献 63
Memon A R,Wang H S,Hussain A.Loop closure detection using supervised and unsupervised deep neural networks for monocular SLAM systems[J].Robotics and Autonomous Systems,2020,126:103470
参考文献 64
Li D J,Shi X S,Long Q W,et al.DXSLAM:a robust and efficient visual SLAM system with deep features[C]//2020 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).October 24,2020-January 24,2021,Las Vegas,NV,USA.IEEE,2020:4958-4965
参考文献 65
Qin C,Zhang Y Z,Liu Y D,et al.Semantic loop closure detection based on graph matching in multi-objects scenes[J].Journal of Visual Communication and Image Representation,2021,76:103072
参考文献 66
Godard C,Aodha O M,Brostow G J.Unsupervised monocular depth estimation with left-right consistency[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6602-6611
参考文献 67
Sang H R,Jiang R,Wang Z P,et al.A novel neural multi-store memory network for autonomous visual navigation in unknown environment[J].IEEE Robotics and Automation Letters,2022,7(2):2039-2046
参考文献 68
Li G H,Chen S L.Visual slam in dynamic scenes based on object tracking and static points detection[J].Journal of Intelligent & Robotic Systems,2022,104(2):1-10
参考文献 69
Liu L,Tang T H,Chen J,et al.Real-time 3D reconstruction using point-dependent pose graph optimization framework[J].Machine Vision and Applications,2022,33(2):1-11
参考文献 70
Xue F,Wang Q,Wang X,et al.Guided feature selection for deep visual odometry[C]//14th Asian Conference on Computer Vision.December 2-6,2018,Perth,Australia.IEEE,2018:293-308
参考文献 71
Tang Y F,Wei C C,Cheng S L,et al.Stereo visual-inertial odometry using structural lines for localizing indoor wheeled robots[J].Measurement Science and Technology,2022,33(5):055114
参考文献 72
Bucci A,Zacchini L,Franchi M,et al.Comparison of feature detection and outlier removal strategies in a mono visual odometry algorithm for underwater navigation[J].Applied Ocean Research,2022,118:102961
参考文献 73
Wu J F,Xiong J,Guo H.Improving robustness of line features for VIO in dynamic scene[J].Measurement Science and Technology,2022,33(6):065204
参考文献 74
Huang W B,Wan W W,Liu H.Optimization-based online initialization and calibration of monocular visual-inertial odometry considering spatial-temporal constraints[J].Sensors(Basel,Switzerland),2021,21(8):2673
参考文献 75
Xiang Y,Fox D.DA-RNN:semantic mapping with data associated recurrent neural networks[J].arXiv e-print,2017,arXiv:1703.03098
目录contents

    摘要

    传统的基于视觉的SLAM技术成果颇丰,但在具有挑战性的环境中难以取得想要的效果.深度学习推动了计算机视觉领域的快速发展,并在图像处理中展现出愈加突出的优势.将深度学习与基于视觉的SLAM结合是一个热门话题,诸多研究人员的努力使二者的广泛结合成为可能.本文从深度学习经典的神经网络入手,介绍了深度学习与传统基于视觉的SLAM算法的结合,概述了卷积神经网络(CNN)与循环神经网络(RNN)在深度估计、位姿估计、闭环检测等方面的成就,分析了神经网络在语义信息提取方面的优点,以期为未来自主移动机器人真正自主化提供帮助.最后,对未来VSLAM发展进行了展望.

    Abstract

    Although traditional vision-based SLAM (VSLAM) technologies have achieved impressive results,they are less satisfactory in challenging environments.Deep learning promotes the rapid development of computer vision and shows prominent advantages in image processing.It's a hot spot to combine deep learning with VSLAM,which is promising through the efforts of many researchers.Here,we introduce the combination of deep learning and traditional VSLAM algorithm,starting from the classical neural networks of deep learning.The achievements of Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) in depth estimation,pose estimation and closed-loop detection are summarized.The advantages of neural network in semantic information extraction are elaborated,and the future development of VSLAM is also prospected.

  • 0 引言

  • 移动机器人执行任务的首要前提是确定自己在所在环境中的位置[1].室外空旷环境下,基于GPS的定位方法可以基本满足机器人的定位需求,但有时接收不到GPS信号[2].室内环境中,通常需要提前设立导航信标如二维码、磁条等,这大大限制了移动机器人的应用范围[3].大多数情况下,移动机器人需要自主完成某些任务,这就要求机器人可以适应足够陌生的环境.因此,能够在未知环境中进行定位和地图构建的SLAM(Simultaneous Localization and Mapping)[4]技术成为自主移动机器人必备的能力.根据所使用的传感器不同,SLAM技术主要分为激光SLAM与视觉SLAM(VSLAM)[5].与激光SLAM相比,VSLAM与人眼类似,主要以图像作为环境感知信息源,更符合人类的认知.近年来,由于相机具有廉价、易安装、可以获得丰富的环境信息、易与其他传感器融合等优势[6],基于相机的VSLAM研究受到了科研人员的广泛关注,大量以视觉为基础的SLAM算法应运而生[7].

  • 随着深度学习的快速发展,不少学者尝试采用深度学习的方法解决视觉SLAM所遇到的问题.深度学习可以根据具体问题学习更强大和有效的特征,并成功地展示了一些具有挑战性的认知和感知任务的良好能力.最近的工作尝试包括从单目图像中对场景进行深度估计,以及视觉里程计和语义映射生成等.权美香等[8]对传统的VSLAM进行了详细总结,并对比了不同方法的优缺点;胡凯等[9]从视觉里程计的角度,对VSLAM的发展做了概述,并介绍了深度学习在VSLAM中的应用;刘瑞军等[10]从里程计、闭环检测等方面介绍了深度学习与VSLAM的结合,并与传统方法进行了对比;李少朋等[11]将基于深度学习的VSLAM与传统的VSLAM进行了对比,并展望了未来发展方向.上述文献大多仅从深度学习角度讲述部分方法,未详细介绍典型神经网络与传统VSLAM的结合,也未将整个发展脉络完整展开.本文首先概述了VSLAM发展脉络,然后从深度学习的两个主要的神经网络,即卷积神经网络(Convolutional Neural Network,CNN)与循环神经网络(Recurrent Neural Network,RNN)入手,重点阐述了神经网络在VSLAM系统中深度估计、位姿估计、闭环检测,以及数据融合等方面的贡献,并介绍了神经网络在语义信息提取方面的优势,最后对VSLAM的发展做出总结和展望.

  • 本文具体结构如下:第1节介绍深度学习中两个典型的神经网络CNN和RNN,并列举了部分优秀的VSLAM算法;第2节阐述了CNN与VSLAM的结合,并从单目深度估计、位姿估计、闭环检测3个方面详细总结了VSLAM的发展进程;第3节重点介绍了RNN与视觉惯性数据融合方面的优势,并给出了神经网络与传统VSLAM结合的部分优秀方案;第4节为总结,并对未来VSLAM的发展做出了展望.

  • 1 神经网络与VSLAM概述

  • 传统的VSLAM研究已经取得了诸多令人惊叹的成就.2007年,Davidson等[12]提出了首个实时的单目VSLAM算法——MonoSLAM,该算法可实现实时无漂移的运动结构恢复.2011年,Newcombe等[13]提出了DTAM算法,该算法被认为是第一个实际意义上的直接法VSLAM.2015年,Mur-Artal等[14]提出了ORB-SLAM算法,创新地使用跟踪、局部建图和闭环检测3个线程同时进行,有效地降低了累计误差.闭环检测线程采用词袋模型BoW[15]进行闭环的检测和修正,在处理速度和构建地图的精度上都取得了很好的效果.随后几年,Mur-Artal团队相继推出了ORB-SLAM2[16]与ORB-SLAM3[17].ORB-SLAM系列是基于特征点提取方法中的佼佼者,它将传统VSLAM方法发展到了十分完善的程度.2018年,Engel等[18]提出了可以有效利用任何图像像素的DSO算法,它是直接法中的经典,其在无特征的区域中也具有良好的鲁棒性,并得到了广泛使用.2018年,香港科技大学团队推出了单目惯性紧耦合的VINS-Mono[19]算法,该算法是视觉惯性融合SLAM中最优秀的算法之一,它充分利用惯性测量单元(Inertial Measurement Unit,IMU)与单目相机的互补性,改善了具有挑战性环境中的定位精度.表1根据前端所用传感器不同,从视觉里程计(Visual Odometry,VO)及视觉惯性里程计(Visual-Inertial Odometry,VIO)两方面列举了部分优秀的传统VSLAM方案,并给出了其开源地址.

  • 传统方法多采用基于特征提取的间接法或者直接对像素进行操作的直接法.虽然在大多数环境中传统方法可以稳定运行,但是在光照强烈、相机快速旋转或是动态物体普遍存在等环境中鲁棒性会大大降低,甚至可能会失效.近年来,深度学习的快速发展吸引了诸多学者的目光,将深度学习的方法与传统VSLAM相结合成为广受关注的研究领域[20].

  • 深度学习可以学习不同数据中的特征或者是数据之间的某种关联,学习得到的特征属性与关联关系都可以用于不同的任务中[29].深度学习通过层次化的处理方式,对视觉数据进行学习,得到数据的抽象表达,在图像识别、语义理解、图像匹配、三维重建[30]等任务中取得了显著的成果[31].作为深度学习中两个重要的神经网络,CNN与RNN在多个领域取得了很高的成就,图1为CNN和RNN的基本框图,表2中给出了两者主要特点的对比.CNN可以从图像中捕捉空间特征,准确地识别物体以及它与图像中其他物体的关系[32].RNN可以有效地处理图像或数值数据,并且由于网络本身具有记忆能力,因此可以学习具有前后相关的数据类型[33].此外,其他类型的神经网络如深度神经网络(Deep Neural Network,DNN),在VSLAM领域也有一些尝试性的工作,但尚在起步阶段.如表3所示,结合深度学习进行VSLAM 的研究已经有了许多突破性的进展.部分学者建议使用深度学习的方法替换传统SLAM的某些模块,如深度估计、闭环检测、位姿估计等,从而改善传统方法.这些方法都取得了一定效果,在不同程度上提高了传统方法的性能.后文将从CNN和RNN两个神经网络入手,重点讲述它们与传统VSLAM的结合.

  • 表1 部分优秀的传统VSLAM算法

  • Table1 Some excellent traditional VSLAM algorithms

  • 图1 CNN和RNN结构

  • Fig.1 CNN and RNN structures

  • 表2 CNN与RNN主要特点对比

  • Table2 Comparison between CNN and RNN

  • 2 CNN与VSLAM

  • CNN以一定的模型对事物进行特征提取,而后根据特征对该事物进行分类、识别、预测或决策等,可以对VSLAM的不同模块提供帮助.

  • 2.1 单目深度估计

  • 基于单目相机的VSLAM算法由于传感器成本低、简单实用,受到了诸多学者的喜爱.单目相机只能得到二维的平面图像,无法获得深度信息.简单地说,单目的局限性主要在于无法得到确定尺度[53] .CNN在图像处理方面的优势已得到充分验证,使用CNN进行视觉深度估计,最大程度上解决了单目相机无法得到可靠的深度信息的问题[54].

  • 2017年,Tateno等[34]在 LSD-SLAM的框架上提出了基于CNN的实时SLAM算法CNN-SLAM.该算法用CNN做深度预测将其输入到后续的传统位姿估计等模块,用来提升定位和建图精度.此外,该算法利用CNN提取环境的语义信息,进行全局地图和语义标签的融合,提高了机器人的环境感知能力.类似利用CNN预测深度信息的工作还有 Code-SLAM[36]以及DVSO[37]等.但上述方法只在某个方面利用了CNN的优势,Yang等[42]提出的 D3VO则从3个层面利用了CNN,包括利用深度学习进行深度估计、位姿估计以及不确定度估计(Σ).如图2所示,D3VO 将预测深度(D)、位姿(Tt-1t)以及不确定度紧密结合到一个直接视觉里程计中,来同时提升前端追踪以及后端非线性优化的性能.所提出的单目深度估计网络的核心是自监督训练体制,这种自监督训练是通过最小化时间立体图像和静态立体图像之间的光度重投影误差来实现的,原理如下:

  • 图2 D3VO核心流程[42]

  • Fig.2 Core processes of D3VO[42]

  • 表3 部分优秀的神经网络与VSLAM结合的算法

  • Table3 Some excellent algorithms combining neural network with VSLAM

  • Lself =1|V|PV mint' rIt,It't.
    (1)
  • 其中:V是图片It上面所有像素的集合,文中将It设置为双目相机中左侧摄像头所得帧;t′是所有源帧的索引(区别于时刻t的某一时刻,右上角的′表示将其与t区分开);It′为包含相邻时间的两帧以及右侧摄像头所得帧,即It′It-1It+1ItsIt-1t时刻前一时刻左侧相机所得帧,It+1t时刻后一时刻左侧相机所得帧,Its为双目相机中右侧摄像头所得帧).

  • 2.2 位姿估计

  • 传统的位姿估计方法,一般采用基于特征的方法或直接法,通过多视图几何来确定相机位姿.但基于特征的方法需要复杂的特征提取和运算[55],直接法则依赖于像素强度值,这使得传统方法在光照强烈或纹理稀疏等环境中很难取得想要的结果[56].基于深度学习的方法由于无需提取环境特征,也无需进行特征匹配和复杂的几何运算,因此更加直观简洁[57].Zhu等[58]通过利用CNN关注光流输入的不同象限来学习旋转和平移,在数据集中测试结果比传统SLAM效果更好.表4 给出了在位姿估计方面传统方法与基于CNN方法的不同.由于CNN的特征检测层通过训练数据进行学习,所以在使用CNN时,避免了显示的特征抽取,而隐式地从训练数据中进行学习,文献[3259]在这方面做出了较为详细的总结.相比传统位姿估计方法,CNN无需传统方法复杂的公式计算,无需提取和匹配特征,因此在线运算速度较快[60].

  • 表4 CNN用于VSLAM与传统方法对比

  • Table4 Comparison of VSLAM algorithms using CNN and traditional methods

  • 2.3 闭环检测

  • 闭环检测可以消除累积轨迹误差和地图误差,决定着整个系统的精度,其本质是场景识别问题[61].在闭环检测方面,传统方法多以词袋模型为基础.如图3所示,首先需要从图像中提取出相互独立的视觉词汇,通常经过特征检测、特征表示以及单词本的生成3个步骤,然后再将新采集到的图像进行词典匹配并分类,过程复杂.而深度学习的强大识别能力,可以提取图像更高层次的稳健特征如语义信息,使得系统能对视角、光照等图像变化具备更强的适应能力,提高闭环图像识别能力[62].因此,基于深度学习的场景识别可以提高闭环检测准确率,CNN用于闭环检测也得到了诸多可靠的结果.

  • Memon等[63]提出一种基于词典的深度学习方法,它不同于传统的BoW词典,创新地使用两个CNN网络一起工作,以加快闭环检测的速度,并忽略移动对象对闭环检测的影响.其核心如图4所示,该方法使用并行线程(标记为虚线框)使闭合检测可以达到更高的速度.将patch逐个送入移动对象识别层,从标记为静止的patch中提取CNN特征,由创新检测层进一步处理.所有不包含任何移动物体的patch再经过创新检测层处理来判断是否访问过该场景.在新的场景下,自动编码器在一个单独的线程上并行地训练这些特征.该方法可以鲁棒地执行循环闭环检测,比同类方法拥有更快的运行速度.Li等[64]使用CNN从每帧图像中提取局部特征和全局特征,然后将这些特征输入现代SLAM模块,用于姿势跟踪、局部映射和重新定位.与传统的基于BoW的方法相比,它的计算效率更高,并且计算成本更低.Qin等[65]采用CNN提取环境语义信息,并将视觉场景建模为语义子图.该方法只保留目标检测中的语义和几何信息,并在数据集中与传统方法进行了比较.结果表明,基于深度学习的特征表示方法,在不提取视觉特征的情况下,可以明显改善闭环检测的效果.

  • 图3 传统的基于视觉词袋模型的图像分类系统结构

  • Fig.3 Structure of traditional image classification system based on bag of visual word model

  • 图4 基于CNN的视觉词袋模型回环检测方法[63]

  • Fig.4 Closed-loop detection using bag of visual word model based on CNN[63]

  • 上述内容主要从单目深度估计、位姿估计、闭环检测3个方面列举了CNN与VSLAM的结合.表5给出了传统方法与结合深度学习方法的对比.CNN在取代传统的特征提取环节上取得了不错的效果,改善了传统特征提取环节消耗时间和计算资源的缺点.同时,CNN有效地提高了单目深度估计的精度.此外,文献[3466]利用CNN提取环境的语义信息,以更高层次的特征来优化传统VSLAM的进程,使得传统VSLAM获得了更好的效果.采用神经网络提取语义信息,并与VSLAM结合将会是一个备受关注的领域,借助语义信息将数据关联从传统的像素级别提升到物体级别,将感知的几何环境信息赋以语义标签,进而得到高层次的语义地图,可帮助机器人进行自主环境理解和人机交互.

  • 3 RNN与VSLAM

  • 循环神经网络RNN的研究从20世纪八九十年代开始,并在21世纪初发展为深度学习经典算法之一,其中长短期记忆网络(Long Short-Term Memory,LSTM)是最常见的循环神经网络之一.LSTM 是 RNN 的一种变体,它记忆可控数量的前期训练数据,或是以更适当的方式遗忘[67].LSTM基本结构如图5所示,从左到右依次为遗忘门、输入门、输出门.采用了特殊隐式单元的LSTM可以长期保存输入,LSTM的这种结构继承了RNN模型的大部分特性,同时解决了梯度反传过程由于逐步缩减而产生的问题.此外GRU(Gate Recurrent Unit)相比LSTM,更容易进行训练,能够很大程度上提高训练效率,因此很多时候会倾向于使用GRU,但在VSLAM领域还处于尝试阶段.

  • 图5 LSTM基本框架包含输入门、输出门、遗忘门

  • Fig.5 The input gate, output gate, and forget gate in basic LSTM framework

  • 遗忘门状态方程为

  • ft=σwf*ht-1,xt+bf.
    (2)
  • 输入门状态方程为

  • it=σwi*ht-1,xt+bi,kt=tanhwk*ht-1,xt+bk.
    (3)
  • 输出门状态方程为

  • Ot=σwo*ht-1,xt+bo,ht=Ot*tanhCt
    (4)
  • 更新后的细胞状态为

  • Ct=ftCt-1+itkt.
    (5)
  • 表5 结合深度学习的VSLAM与传统方法对比

  • Table5 Comparison of VSLAM algorithms combined with deep learning and traditional methods

  • 循环神经网络具有记忆性,参数共享,因此,在对序列的非线性特征进行学习时具有一定优势.RNN在帮助建立相邻帧之间的一致性方面具有很大的优势,高层特征具备更好的区分性,可以帮助机器人更好地完成数据关联.

  • 3.1 位姿估计

  • 传统的位姿估计方法首先需要特征提取与匹配[68],或是基于像素亮度变化的复杂计算.其原理如图6所示,该问题的核心是求解旋转矩阵和平移向量,需要繁琐的计算过程.基于特征的方法(图6a)需要十分耗时地提取特征,计算描述子的操作丢失了除了特征点以外的很多信息(图6a中Rt分别为旋转矩阵和平移向量,红色点为空间中的特征点,黑色点为特征点在不同图像中的投影).直接法(图6b)不同于特征点法最小化重投影误差,而是通过最小化相邻帧之间的灰度误差估计相机运动,但是基于灰度不变假设:

  • I(x,y,z)=I(x+Δx,y+Δy,z+Δz).
    (6)
  • 如图6b,假设空间点P在相邻两帧图像上的投影分别为P1P2两点(用不同颜色的点表示二者像素强度的差别).它们的像素强度分别为I1P1,i)和I2P2,i),其中,i表示当前图像中第i个点.则优化目标就是这两点的亮度误差ei的二范数.

  • minξ J(ξ)=i=1N eiTei
    (7)
  • ei=I1P1,i-I2P2,i,
    (8)
  • P1,i=TP2,i,
    (9)
  • T=expξ
    (10)
  • 其中,Tξ分别是P1P2之间的转换矩阵及其李代数.ξ右上角的∧表示把ξ转为一个四维矩阵,从而通过指数映射成为变换矩阵.

  • 通过引入端对端的深度学习方法,使得视觉图像帧间的位姿参数解算无须特征匹配与复杂的几何运算,可快速得到帧间相对位姿参数[69].Xue等[70]基于RNN来实现位姿的估计.在位姿估计过程中,旋转和位移是分开进行训练的,相对于传统方法有更好的适应性.2017年,Wang等[35]使用深度递归卷积神经网络,提出一种新颖的端到端单目VO的框架.由于它是以端到端的方式进行训练和配置的,因此可以直接从一系列原始的RGB图像中计算得到姿态,而无需采用任何传统VO框架中的模块.该方法做到了视觉里程计的端到端实现,免去了帧间各种几何关系的约束计算,有良好的泛化能力.如图7所示,该方案使用CNN+RNN对相机的运动进行估计,直接从原始RGB图像序列推断姿态.它不仅通过卷积神经网络自动学习VO问题的有效特征表示,而且还利用深度回归神经网络隐式建模顺序动力学和关系.

  • 3.2 视觉惯性融合

  • 由于惯性测量元件IMU能够在短时间内高频地获得精准的估计,减轻动态物体对相机的影响,而相机数据也能有效地修正IMU的累积漂移,IMU被认为是与相机互补性最强的传感器之一[71].传统方法中,视觉惯性融合按照是否将图像特征信息加入到状态向量中可以分为松耦合和紧耦合[72].松耦合是指IMU和相机分别进行自身的运动估计,然后对其位姿估计输出结果进行融合[73].紧耦合是指把IMU的状态与相机的状态合并在一起,共同构建运动方程和观测方程,然后进行状态估计[74].图8为传统方法典型的视觉惯性融合流程,由于相机和IMU频率相差较大,需要先进行严格的同步校准.但是,不同传感器的数据融合,势必会带来计算资源消耗过多、实时性差等问题.

  • RNN是深度学习领域数据驱动的时序建模的常用方法,IMU输出的高帧率角速度、加速度等惯性数据,在时序上有着严格的依赖关系,特别适合使用RNN这类模型来优化.Clark等[47]使用一个常规的小型LSTM网络来处理IMU的原始数据,得到了IMU数据下的运动特征.如图9所示,在对相机数据和IMU数据结合后,送入一个核心的LSTM网络进行特征融合和位姿估计.该方法通过神经网络方法,避免了传统方法复杂的数据融合过程,使得运行效率大大提升.

  • 图6 基于特征法的提取与匹配(a)和基于直接法的像素强度变化计算(b)

  • Fig.6 (a) Extraction and matching based on feature method, and (b) calculation of pixel intensity change based on direct method

  • 图7 CNN与RNN结合用来改善传统VO[35]

  • Fig.7 Traditional VO improved via combination of CNN and RNN[35]

  • 图8 传统的视觉惯性融合流程

  • Fig.8 Traditional visual-inertial fusion process

  • 相比于单纯用于位姿估计, RNN在视觉惯性数据融合方面做出的贡献更具吸引力.此类方法对视觉惯性数据进行了非常有效的融合,相比传统方法更便捷,类似的工作有文献[50-51]等.此外,一些工作利用神经网络提取环境中的语义信息,高层特征更具区分性,对于VSLAM数据关联有很好的帮助.2017年,Xiang等[75]使用RNN与KinectFusion相结合,对RGB-D相机采集图像进行语义标注,用来重建三维语义地图.通过在RNN中引入了一个新的循环单元,来解决GPU计算资源消耗过大的问题.该方法充分利用RNN的优点,实现了语义信息的标注,高层特征具备更好的区分性,同时帮助机器人更好地完成数据关联.

  • 4 总结与展望

  • 本文对深度学习中的两个典型神经网络CNN与RNN进行了介绍,并详细总结了神经网络在VSLAM中的贡献,从深度估计、位姿估计、闭环检测等方面将基于神经网络的方法与传统方法进行对比.从CNN与RNN各自的特点入手,列举出其对传统VSLAM不同模块的改善.神经网络一定程度上改善了传统VSLAM由于设计特征而带来的应用局限性,同时对高层语义快速准确生成以及机器人知识库构建也产生了重要影响,从而提高了机器人的学习能力和智能化水平.

  • 图9 采用CNN+LSTM的神经网络结构[47]

  • Fig.9 Neural network structure of CNN+LSTM[47]

  • 综合他人所做研究,笔者认为未来VSLAM的发展趋势如下:

  • 1)更高层次的环境感知.神经网络可以更加方便地提取环境中高层次的语义信息,可以促进机器人智能化的发展.传统的VSLAM算法只能满足机器人基本的定位导航需求,无法完成更高级别的任务,如“帮我把卧室门关上”、“去厨房帮我拿个苹果”等.借助语义信息将数据关联从传统的像素级别提升到物体级别,将感知的几何环境信息赋以语义标签,进而得到高层次的语义地图,可以帮助机器人进行自主环境理解和人机交互,实现真正自主化.

  • 2)更完善的理论支撑体系.通过深度学习技术学习的信息特征还缺少直观的意义以及清晰的理论指导.目前深度学习多应用于SLAM局部的子模块,如深度估计、闭环检测等,而如何将深度学习应用贯穿于整个 SLAM 系统仍是一个巨大挑战.

  • 3)更高效的数据融合.CNN可以与VLSAM的诸多环节进行结合,如特征提取与匹配、深度估计、位姿估计等, RNN的应用范围较小.但RNN在数据融合方面的优势,可以更好地融合多传感器的数据,快速推动传感器融合SLAM技术的发展.未来可能会更多地关注CNN与RNN的结合,来提升VSLAM的整体性能.

  • 参考文献

    • [1] 任伟建,高强,康朝海,等.移动机器人同步定位与建图技术综述[J].计算机测量与控制,2022,30(2):1-10,37.REN Weijian,GAO Qiang,KANG Chaohai,et al.Overview of simultaneous localization and mapping technology of mobile robots[J].Computer Measurement & Control,2022,30(2):1-10,37

    • [2] 赵乐文,任嘉倩,丁杨.基于GNSS的空间环境参数反演平台及精度评估[J].南京信息工程大学学报(自然科学版),2021,13(2):204-210.ZHAO Lewen,REN Jiaqian,DING Yang.Platform for GNSS real-time space environment parameter inversion and its accuracy evaluation[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2021,13(2):204-210

    • [3] 尹姝,陈元橼,仇翔.基于RFID和自适应卡尔曼滤波的室内移动目标定位方法[J].南京信息工程大学学报(自然科学版),2018,10(6):749-753.YIN Shu,CHEN Yuanyuan,QIU Xiang.Indoor moving-target localization using RFID and adaptive Kalman filter[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2018,10(6):749-753

    • [4] 周韦,孙宪坤,吴飞.基于SLAM/UWB的室内融合定位算法研究[J].全球定位系统,2022,47(1):36-42,85.ZHOU Wei,SUN Xiankun,WU Fei.Research on indoor fusion positioning algorithm based on SLAM/UWB[J].GNSS World of China,2022,47(1):36-42,85

    • [5] Bresson G,Alsayed Z,Yu L,et al.Simultaneous localization and mapping:a survey of current trends in autonomous driving[J].IEEE Transactions on Intelligent Vehicles,2017,2(3):194-220

    • [6] 李晓飞,宋亚男,徐荣华,等.基于双目视觉的船舶跟踪与定位[J].南京信息工程大学学报(自然科学版),2015,7(1):46-52.LI Xiaofei,SONG Yanan,XU Ronghua,et al.Tracking and positioning of ship based on binocular vision[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2015,7(1):46-52

    • [7] 刘明芹,张晓光,徐桂云,等.单机器人SLAM技术的发展及相关主流技术综述[J].计算机工程与应用,2020,56(18):25-35.LIU Mingqin,ZHANG Xiaoguang,XU Guiyun,et al.Review of development of single robot SLAM technology and related mainstream technology[J].Computer Engineering and Applications,2020,56(18):25-35

    • [8] 权美香,朴松昊,李国.视觉SLAM综述[J].智能系统学报,2016(6):768-776.QUAN Meixiang,PIAO Songhao,LI Guo.An overview of visual SLAM[J].CAAI Transactions on Intelligent Systems,2016(6):768-776

    • [9] 胡凯,吴佳胜,郑翡,等.视觉里程计研究综述[J].南京信息工程大学学报(自然科学版),2021,13(3):269-280.HU Kai,WU Jiasheng,ZHENG Fei,et al.A survey of visual odometry[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2021,13(3):269-280

    • [10] 刘瑞军,王向上,张晨,等.基于深度学习的视觉SLAM综述[J].系统仿真学报,2020,32(7):1244-1256.LIU Ruijun,WANG Xiangshang,ZHANG Chen,et al.A survey on visual SLAM based on deep learning[J].Journal of System Simulation,2020,32(7):1244-1256

    • [11] 李少朋,张涛.深度学习在视觉SLAM中应用综述[J].空间控制技术与应用,2019,45(2):1-10.LI Shaopeng,ZHANG Tao.A survey of deep learning application in visual SLAM[J].Aerospace Control and Application,2019,45(2):1-10

    • [12] Davison A J,Reid I D,Molton N D,et al.MonoSLAM:real-time single camera SLAM[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2007,29(6):1052-1067

    • [13] Newcombe R A,Lovegrove S J,Davison A J.DTAM:dense tracking and mapping in real-time[C]//2011 International Conference on Computer Vision.November 6-13,2011,Barcelona,Spain.IEEE,2011:2320-2327

    • [14] Mur-Artal R,Montiel J M M,Tardos J D.ORB-SLAM:a versatile and accurate monocular SLAM system[J].IEEE Transactions on Robotics,2015,31(5):1147-1163

    • [15] Galvez-Lopez D,Tardos J D.Bags of binary words for fast place recognition in image sequences[J].IEEE Transactions on Robotics,2012,28(5):1188-1197

    • [16] Mur-Artal R,Tardos J D.ORB-SLAM2:an open-source SLAM system for monocular,stereo,and RGB-D cameras[J].IEEE Transactions on Robotics,2017,33(5):1255-1262

    • [17] Campos C,Elvira R,Rodriguez J J G,et al.ORB-SLAM3:an accurate open-source library for visual,visual-inertial,and multimap SLAM[J].IEEE Transactions on Robotics,2021,37(6):1874-1890

    • [18] Engel J,Koltun V,Cremers D.Direct sparse odometry[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(3):611-625

    • [19] Qin T,Li P L,Shen S J.VINS-mono:a robust and versatile monocular visual-inertial state estimator[J].IEEE Transactions on Robotics,2018,34(4):1004-1020

    • [20] 邓晨,李宏伟,张斌,等.基于深度学习的语义SLAM关键帧图像处理[J].测绘学报,2021,50(11):1605-1616.DENG Chen,LI Hongwei,ZHANG Bin,et al.Research on key frame image processing of semantic SLAM based on deep learning[J].Acta Geodaetica et Cartographica Sinica,2021,50(11):1605-1616

    • [21] Klein G,Murray D.Parallel tracking and mapping for small AR workspaces[C]//2007 6th IEEE and ACM International Symposium on Mixed and Augmented Reality.November 13-16,2017,Nara,Japan.IEEE,2007:225-234

    • [22] Gomez-Ojeda R,Briales J,Gonzalez-Jimenez J.PL-SVO:semi-direct monocular visual odometry by combining points and line segments[C]//2016 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).October 9-14,2016,Daejeon,Korea(South).IEEE,2016:4211-4216

    • [23] Pumarola A,Vakhitov A,Agudo A,et al.PL-SLAM:real-time monocular visual SLAM with points and lines[C]//2017 IEEE International Conference on Robotics and Automation.May 29-June 3,2017,Singapore.IEEE,2017:4503-4508

    • [24] Forster C,Pizzoli M,Scaramuzza D.SVO:fast semi-direct monocular visual odometry[C]//2014 IEEE International Conference on Robotics and Automation.May 31-June 7,2014,Hong Kong,China.IEEE,2014:15-22

    • [25] Engel J,Schps T,Cremers D.LSD-SLAM:large-scale direct monocular SLAM[C]//European Conference on Computer Vision.Springer,Cham,2014:834-849

    • [26] Mourikis A I,Roumeliotis S I.A multi-state constraint Kalman filter for vision-aided inertial navigation[C]//Proceedings 2007 IEEE International Conference on Robotics and Automation.April 10-14,2007,Rome,Italy.IEEE,2007:3565-3572

    • [27] Leutenegger S,Lynen S,Bosse M,et al.Keyframe-based visual-inertial odometry using nonlinear optimization[J].The International Journal of Robotics Research,2015,34(3):314-334

    • [28] Bloesch M,Omari S,Hutter M,et al.Robust visual inertial odometry using a direct EKF-based approach[C]//2015 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 28-October 2,2015,Hamburg,Germany.IEEE,2015:298-304

    • [29] Xu D,Vedaldi A,Henriques J F.Moving SLAM:fully unsupervised deep learning in non-rigid scenes[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 27-October 1,2021,Prague,Czech Republic.IEEE,2021:4611-4617

    • [30] 张彦雯,胡凯,王鹏盛.三维重建算法研究综述[J].南京信息工程大学学报(自然科学版),2020,12(5):591-602.ZHANG Yanwen,HU Kai,WANG Pengsheng.Review of 3D reconstruction algorithms[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2020,12(5):591-602

    • [31] Li J L,Li Z J,Feng Y,et al.Development of a human-robot hybrid intelligent system based on brain teleoperation and deep learning SLAM[J].IEEE Transactions on Automation Science and Engineering,2019,16(4):1664-1674

    • [32] Mumuni A,Mumuni F.CNN architectures for geometric transformation-invariant feature representation in computer vision:a review[J].SN Computer Science,2021,2(5):1-23

    • [33] Ma R B,Wang R,Zhang Y B,et al.RNNSLAM:reconstructing the 3D colon to visualize missing regions during a colonoscopy[J].Medical Image Analysis,2021,72:102100

    • [34] Tateno K,Tombari F,Laina I,et al.CNN-SLAM:real-time dense monocular SLAM with learned depth prediction[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6565-6574

    • [35] Wang S,Clark R,Wen H K,et al.DeepVO:towards end-to-end visual odometry with deep recurrent convolutional neural networks[C]//2017 IEEE International Conference on Robotics and Automation.May 29-June 3,2017,Singapore.IEEE,2017:2043-2050

    • [36] Bloesch M,Czarnowski J,Clark R,et al.CodeSLAM:learning a compact,optimisable representation for dense visual SLAM[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.June 18-23,2018,Salt Lake City,UT,USA.IEEE,2018:2560-2568

    • [37] Yang N,Wang R,Stuckler J,et al.Deep virtual stereo odometry:leveraging deep depth prediction for monocular direct sparse odometry[C]//European Conference on Computer Vision(ECCV).September 8-12,2018,Munich,Germany.2018.DOI:10.48550/arXiv.1807.02570

    • [38] Li R H,Wang S,Long Z Q,et al.UnDeepVO:monocular visual odometry through unsupervised deep learning[C]//2018 IEEE International Conference on Robotics and Automation.May 21-25,2018,Brisbane,QLD,Australia.IEEE,2018:7286-7291

    • [39] Loo S Y,Amiri A J,Mashohor S,et al.CNN-SVO:improving the mapping in semi-direct visual odometry using single-image depth prediction[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5218-5223

    • [40] Almalioglu Y,Saputra M R U,de Gusmão P P B,et al.GANVO:unsupervised deep monocular visual odometry and depth estimation with generative adversarial networks[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5474-5480

    • [41] Li Y,Ushiku Y,Harada T.Pose graph optimization for unsupervised monocular visual odometry[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5439-5445

    • [42] Yang N,von Stumberg L,Wang R,et al.D3VO:deep depth,deep pose and deep uncertainty for monocular visual odometry[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).June 13-19,2020,Seattle,WA,USA.IEEE,2020:1278-1289

    • [43] Chancán M,Milford M.DeepSeqSLAM:a trainable CNN+RNN for joint global description and sequence-based place recognition[J].arXiv e-print,2020,arXiv:2011.08518

    • [44] Li R H,Wang S,Gu D B.DeepSLAM:a robust monocular SLAM system with unsupervised deep learning[J].IEEE Transactions on Industrial Electronics,2021,68(4):3577-3587

    • [45] Bruno H M S,Colombini E L.LIFT-SLAM:a deep-learning feature-based monocular visual SLAM method[J].Neurocomputing,2021,455:97-110

    • [46] Zhang S M,Lu S Y,He R,et al.Stereo visual odometry pose correction through unsupervised deep learning[J].Sensors(Basel,Switzerland),2021,21(14):4735

    • [47] Clark R,Wang S,Wen H,et al.VINet:visual-inertial odometry as a sequence-to-sequence learning problem[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence.February 4-9,2017,San Francisco,CA,USA.2017:3995-4001

    • [48] Shamwell E J,Lindgren K,Leung S,et al.Unsuperviseddeep visual-inertial odometry with online error correction for RGB-D imagery[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(10):2478-2493

    • [49] Han L M,Lin Y M,Du G G,et al.DeepVIO:self-supervised deep learning of monocular visual inertial odometry using 3D geometric constraints[C]//2019 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).November 3-8,2019,Macao,China.IEEE,2019:6906-6913

    • [50] Chen C H,Rosa S,Miao Y S,et al.Selective sensor fusion for neural visual-inertial odometry[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).June 15-20,2019,Long Beach,CA,USA.IEEE,2019:10534-10543

    • [51] Kim Y,Yoon S,Kim S,et al.Unsupervised balanced covariance learning for visual-inertial sensor fusion[J].IEEE Robotics and Automation Letters,2021,6(2):819-826

    • [52] Gurturk M,Yusefi A,Aslan M F,et al.The YTU dataset and recurrent neural network based visual-inertial odometry[J].Measurement,2021,184:109878

    • [53] 傅杰,徐常胜.关于单目标跟踪方法的研究综述[J].南京信息工程大学学报(自然科学版),2019,11(6):638-650.FU Jie,XU Changsheng.A survey of single object tracking methods[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2019,11(6):638-650

    • [54] Steenbeek A,Nex F.CNN-based dense monocular visual SLAM for real-time UAV exploration in emergency conditions[J].Drones,2022,6(3):79

    • [55] 唐灿,唐亮贵,刘波.图像特征检测与匹配方法研究综述[J].南京信息工程大学学报(自然科学版),2020,12(3):261-273.TANG Can,TANG Lianggui,LIU Bo.A survey of image feature detection and matching methods[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2020,12(3):261-273

    • [56] Li L,Kong X,Zhao XR,et al.Semantic scan context:a novel semantic-based loop-closure method for LiDAR SLAM[J].Autonomous Robots,2022,46(4):535-551

    • [57] Sakkari M,Hamdi M,Elmannai H,et al.Feature extraction-based deep self-organizing map[J].Circuits,Systems,and Signal Processing,2022,41(5):2802-2824

    • [58] Zhu R,Yang M K,Liu W,et al.DeepAVO:efficient pose refining with feature distilling for deep visual odometry[J].Neurocomputing,2022,467:22-35

    • [59] Kim J J Y,Urschler M,Riddle P J,et al.SymbioLCD:ensemble-based loop closure detection using CNN-extracted objects and visual bag-of-words[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 27-October 1,2021,Prague,Czech Republic.IEEE,2021:5425

    • [60] Ai Y B,Rui T,Lu M,et al.DDL-SLAM:a robust RGB-D SLAM in dynamic environments combined with deep learning[J].IEEE Access,8:162335-162342

    • [61] Javed Z,Kim G W.PanoVILD:a challenging panoramic vision,inertial and LiDAR dataset for simultaneous localization and mapping[J].The Journal of Supercomputing,2022,78(6):8247-8267

    • [62] Duan R,Feng Y R,Wen C Y.Deep pose graph-matching-based loop closure detection for semantic visual SLAM[J].Sustainability,2022,14(19):11864

    • [63] Memon A R,Wang H S,Hussain A.Loop closure detection using supervised and unsupervised deep neural networks for monocular SLAM systems[J].Robotics and Autonomous Systems,2020,126:103470

    • [64] Li D J,Shi X S,Long Q W,et al.DXSLAM:a robust and efficient visual SLAM system with deep features[C]//2020 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).October 24,2020-January 24,2021,Las Vegas,NV,USA.IEEE,2020:4958-4965

    • [65] Qin C,Zhang Y Z,Liu Y D,et al.Semantic loop closure detection based on graph matching in multi-objects scenes[J].Journal of Visual Communication and Image Representation,2021,76:103072

    • [66] Godard C,Aodha O M,Brostow G J.Unsupervised monocular depth estimation with left-right consistency[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6602-6611

    • [67] Sang H R,Jiang R,Wang Z P,et al.A novel neural multi-store memory network for autonomous visual navigation in unknown environment[J].IEEE Robotics and Automation Letters,2022,7(2):2039-2046

    • [68] Li G H,Chen S L.Visual slam in dynamic scenes based on object tracking and static points detection[J].Journal of Intelligent & Robotic Systems,2022,104(2):1-10

    • [69] Liu L,Tang T H,Chen J,et al.Real-time 3D reconstruction using point-dependent pose graph optimization framework[J].Machine Vision and Applications,2022,33(2):1-11

    • [70] Xue F,Wang Q,Wang X,et al.Guided feature selection for deep visual odometry[C]//14th Asian Conference on Computer Vision.December 2-6,2018,Perth,Australia.IEEE,2018:293-308

    • [71] Tang Y F,Wei C C,Cheng S L,et al.Stereo visual-inertial odometry using structural lines for localizing indoor wheeled robots[J].Measurement Science and Technology,2022,33(5):055114

    • [72] Bucci A,Zacchini L,Franchi M,et al.Comparison of feature detection and outlier removal strategies in a mono visual odometry algorithm for underwater navigation[J].Applied Ocean Research,2022,118:102961

    • [73] Wu J F,Xiong J,Guo H.Improving robustness of line features for VIO in dynamic scene[J].Measurement Science and Technology,2022,33(6):065204

    • [74] Huang W B,Wan W W,Liu H.Optimization-based online initialization and calibration of monocular visual-inertial odometry considering spatial-temporal constraints[J].Sensors(Basel,Switzerland),2021,21(8):2673

    • [75] Xiang Y,Fox D.DA-RNN:semantic mapping with data associated recurrent neural networks[J].arXiv e-print,2017,arXiv:1703.03098

  • 参考文献

    • [1] 任伟建,高强,康朝海,等.移动机器人同步定位与建图技术综述[J].计算机测量与控制,2022,30(2):1-10,37.REN Weijian,GAO Qiang,KANG Chaohai,et al.Overview of simultaneous localization and mapping technology of mobile robots[J].Computer Measurement & Control,2022,30(2):1-10,37

    • [2] 赵乐文,任嘉倩,丁杨.基于GNSS的空间环境参数反演平台及精度评估[J].南京信息工程大学学报(自然科学版),2021,13(2):204-210.ZHAO Lewen,REN Jiaqian,DING Yang.Platform for GNSS real-time space environment parameter inversion and its accuracy evaluation[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2021,13(2):204-210

    • [3] 尹姝,陈元橼,仇翔.基于RFID和自适应卡尔曼滤波的室内移动目标定位方法[J].南京信息工程大学学报(自然科学版),2018,10(6):749-753.YIN Shu,CHEN Yuanyuan,QIU Xiang.Indoor moving-target localization using RFID and adaptive Kalman filter[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2018,10(6):749-753

    • [4] 周韦,孙宪坤,吴飞.基于SLAM/UWB的室内融合定位算法研究[J].全球定位系统,2022,47(1):36-42,85.ZHOU Wei,SUN Xiankun,WU Fei.Research on indoor fusion positioning algorithm based on SLAM/UWB[J].GNSS World of China,2022,47(1):36-42,85

    • [5] Bresson G,Alsayed Z,Yu L,et al.Simultaneous localization and mapping:a survey of current trends in autonomous driving[J].IEEE Transactions on Intelligent Vehicles,2017,2(3):194-220

    • [6] 李晓飞,宋亚男,徐荣华,等.基于双目视觉的船舶跟踪与定位[J].南京信息工程大学学报(自然科学版),2015,7(1):46-52.LI Xiaofei,SONG Yanan,XU Ronghua,et al.Tracking and positioning of ship based on binocular vision[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2015,7(1):46-52

    • [7] 刘明芹,张晓光,徐桂云,等.单机器人SLAM技术的发展及相关主流技术综述[J].计算机工程与应用,2020,56(18):25-35.LIU Mingqin,ZHANG Xiaoguang,XU Guiyun,et al.Review of development of single robot SLAM technology and related mainstream technology[J].Computer Engineering and Applications,2020,56(18):25-35

    • [8] 权美香,朴松昊,李国.视觉SLAM综述[J].智能系统学报,2016(6):768-776.QUAN Meixiang,PIAO Songhao,LI Guo.An overview of visual SLAM[J].CAAI Transactions on Intelligent Systems,2016(6):768-776

    • [9] 胡凯,吴佳胜,郑翡,等.视觉里程计研究综述[J].南京信息工程大学学报(自然科学版),2021,13(3):269-280.HU Kai,WU Jiasheng,ZHENG Fei,et al.A survey of visual odometry[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2021,13(3):269-280

    • [10] 刘瑞军,王向上,张晨,等.基于深度学习的视觉SLAM综述[J].系统仿真学报,2020,32(7):1244-1256.LIU Ruijun,WANG Xiangshang,ZHANG Chen,et al.A survey on visual SLAM based on deep learning[J].Journal of System Simulation,2020,32(7):1244-1256

    • [11] 李少朋,张涛.深度学习在视觉SLAM中应用综述[J].空间控制技术与应用,2019,45(2):1-10.LI Shaopeng,ZHANG Tao.A survey of deep learning application in visual SLAM[J].Aerospace Control and Application,2019,45(2):1-10

    • [12] Davison A J,Reid I D,Molton N D,et al.MonoSLAM:real-time single camera SLAM[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2007,29(6):1052-1067

    • [13] Newcombe R A,Lovegrove S J,Davison A J.DTAM:dense tracking and mapping in real-time[C]//2011 International Conference on Computer Vision.November 6-13,2011,Barcelona,Spain.IEEE,2011:2320-2327

    • [14] Mur-Artal R,Montiel J M M,Tardos J D.ORB-SLAM:a versatile and accurate monocular SLAM system[J].IEEE Transactions on Robotics,2015,31(5):1147-1163

    • [15] Galvez-Lopez D,Tardos J D.Bags of binary words for fast place recognition in image sequences[J].IEEE Transactions on Robotics,2012,28(5):1188-1197

    • [16] Mur-Artal R,Tardos J D.ORB-SLAM2:an open-source SLAM system for monocular,stereo,and RGB-D cameras[J].IEEE Transactions on Robotics,2017,33(5):1255-1262

    • [17] Campos C,Elvira R,Rodriguez J J G,et al.ORB-SLAM3:an accurate open-source library for visual,visual-inertial,and multimap SLAM[J].IEEE Transactions on Robotics,2021,37(6):1874-1890

    • [18] Engel J,Koltun V,Cremers D.Direct sparse odometry[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(3):611-625

    • [19] Qin T,Li P L,Shen S J.VINS-mono:a robust and versatile monocular visual-inertial state estimator[J].IEEE Transactions on Robotics,2018,34(4):1004-1020

    • [20] 邓晨,李宏伟,张斌,等.基于深度学习的语义SLAM关键帧图像处理[J].测绘学报,2021,50(11):1605-1616.DENG Chen,LI Hongwei,ZHANG Bin,et al.Research on key frame image processing of semantic SLAM based on deep learning[J].Acta Geodaetica et Cartographica Sinica,2021,50(11):1605-1616

    • [21] Klein G,Murray D.Parallel tracking and mapping for small AR workspaces[C]//2007 6th IEEE and ACM International Symposium on Mixed and Augmented Reality.November 13-16,2017,Nara,Japan.IEEE,2007:225-234

    • [22] Gomez-Ojeda R,Briales J,Gonzalez-Jimenez J.PL-SVO:semi-direct monocular visual odometry by combining points and line segments[C]//2016 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).October 9-14,2016,Daejeon,Korea(South).IEEE,2016:4211-4216

    • [23] Pumarola A,Vakhitov A,Agudo A,et al.PL-SLAM:real-time monocular visual SLAM with points and lines[C]//2017 IEEE International Conference on Robotics and Automation.May 29-June 3,2017,Singapore.IEEE,2017:4503-4508

    • [24] Forster C,Pizzoli M,Scaramuzza D.SVO:fast semi-direct monocular visual odometry[C]//2014 IEEE International Conference on Robotics and Automation.May 31-June 7,2014,Hong Kong,China.IEEE,2014:15-22

    • [25] Engel J,Schps T,Cremers D.LSD-SLAM:large-scale direct monocular SLAM[C]//European Conference on Computer Vision.Springer,Cham,2014:834-849

    • [26] Mourikis A I,Roumeliotis S I.A multi-state constraint Kalman filter for vision-aided inertial navigation[C]//Proceedings 2007 IEEE International Conference on Robotics and Automation.April 10-14,2007,Rome,Italy.IEEE,2007:3565-3572

    • [27] Leutenegger S,Lynen S,Bosse M,et al.Keyframe-based visual-inertial odometry using nonlinear optimization[J].The International Journal of Robotics Research,2015,34(3):314-334

    • [28] Bloesch M,Omari S,Hutter M,et al.Robust visual inertial odometry using a direct EKF-based approach[C]//2015 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 28-October 2,2015,Hamburg,Germany.IEEE,2015:298-304

    • [29] Xu D,Vedaldi A,Henriques J F.Moving SLAM:fully unsupervised deep learning in non-rigid scenes[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 27-October 1,2021,Prague,Czech Republic.IEEE,2021:4611-4617

    • [30] 张彦雯,胡凯,王鹏盛.三维重建算法研究综述[J].南京信息工程大学学报(自然科学版),2020,12(5):591-602.ZHANG Yanwen,HU Kai,WANG Pengsheng.Review of 3D reconstruction algorithms[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2020,12(5):591-602

    • [31] Li J L,Li Z J,Feng Y,et al.Development of a human-robot hybrid intelligent system based on brain teleoperation and deep learning SLAM[J].IEEE Transactions on Automation Science and Engineering,2019,16(4):1664-1674

    • [32] Mumuni A,Mumuni F.CNN architectures for geometric transformation-invariant feature representation in computer vision:a review[J].SN Computer Science,2021,2(5):1-23

    • [33] Ma R B,Wang R,Zhang Y B,et al.RNNSLAM:reconstructing the 3D colon to visualize missing regions during a colonoscopy[J].Medical Image Analysis,2021,72:102100

    • [34] Tateno K,Tombari F,Laina I,et al.CNN-SLAM:real-time dense monocular SLAM with learned depth prediction[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6565-6574

    • [35] Wang S,Clark R,Wen H K,et al.DeepVO:towards end-to-end visual odometry with deep recurrent convolutional neural networks[C]//2017 IEEE International Conference on Robotics and Automation.May 29-June 3,2017,Singapore.IEEE,2017:2043-2050

    • [36] Bloesch M,Czarnowski J,Clark R,et al.CodeSLAM:learning a compact,optimisable representation for dense visual SLAM[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.June 18-23,2018,Salt Lake City,UT,USA.IEEE,2018:2560-2568

    • [37] Yang N,Wang R,Stuckler J,et al.Deep virtual stereo odometry:leveraging deep depth prediction for monocular direct sparse odometry[C]//European Conference on Computer Vision(ECCV).September 8-12,2018,Munich,Germany.2018.DOI:10.48550/arXiv.1807.02570

    • [38] Li R H,Wang S,Long Z Q,et al.UnDeepVO:monocular visual odometry through unsupervised deep learning[C]//2018 IEEE International Conference on Robotics and Automation.May 21-25,2018,Brisbane,QLD,Australia.IEEE,2018:7286-7291

    • [39] Loo S Y,Amiri A J,Mashohor S,et al.CNN-SVO:improving the mapping in semi-direct visual odometry using single-image depth prediction[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5218-5223

    • [40] Almalioglu Y,Saputra M R U,de Gusmão P P B,et al.GANVO:unsupervised deep monocular visual odometry and depth estimation with generative adversarial networks[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5474-5480

    • [41] Li Y,Ushiku Y,Harada T.Pose graph optimization for unsupervised monocular visual odometry[C]//2019 International Conference on Robotics and Automation(ICRA).May 20-24,2019,Montreal,QC,Canada.IEEE,2019:5439-5445

    • [42] Yang N,von Stumberg L,Wang R,et al.D3VO:deep depth,deep pose and deep uncertainty for monocular visual odometry[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).June 13-19,2020,Seattle,WA,USA.IEEE,2020:1278-1289

    • [43] Chancán M,Milford M.DeepSeqSLAM:a trainable CNN+RNN for joint global description and sequence-based place recognition[J].arXiv e-print,2020,arXiv:2011.08518

    • [44] Li R H,Wang S,Gu D B.DeepSLAM:a robust monocular SLAM system with unsupervised deep learning[J].IEEE Transactions on Industrial Electronics,2021,68(4):3577-3587

    • [45] Bruno H M S,Colombini E L.LIFT-SLAM:a deep-learning feature-based monocular visual SLAM method[J].Neurocomputing,2021,455:97-110

    • [46] Zhang S M,Lu S Y,He R,et al.Stereo visual odometry pose correction through unsupervised deep learning[J].Sensors(Basel,Switzerland),2021,21(14):4735

    • [47] Clark R,Wang S,Wen H,et al.VINet:visual-inertial odometry as a sequence-to-sequence learning problem[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence.February 4-9,2017,San Francisco,CA,USA.2017:3995-4001

    • [48] Shamwell E J,Lindgren K,Leung S,et al.Unsuperviseddeep visual-inertial odometry with online error correction for RGB-D imagery[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(10):2478-2493

    • [49] Han L M,Lin Y M,Du G G,et al.DeepVIO:self-supervised deep learning of monocular visual inertial odometry using 3D geometric constraints[C]//2019 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).November 3-8,2019,Macao,China.IEEE,2019:6906-6913

    • [50] Chen C H,Rosa S,Miao Y S,et al.Selective sensor fusion for neural visual-inertial odometry[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).June 15-20,2019,Long Beach,CA,USA.IEEE,2019:10534-10543

    • [51] Kim Y,Yoon S,Kim S,et al.Unsupervised balanced covariance learning for visual-inertial sensor fusion[J].IEEE Robotics and Automation Letters,2021,6(2):819-826

    • [52] Gurturk M,Yusefi A,Aslan M F,et al.The YTU dataset and recurrent neural network based visual-inertial odometry[J].Measurement,2021,184:109878

    • [53] 傅杰,徐常胜.关于单目标跟踪方法的研究综述[J].南京信息工程大学学报(自然科学版),2019,11(6):638-650.FU Jie,XU Changsheng.A survey of single object tracking methods[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2019,11(6):638-650

    • [54] Steenbeek A,Nex F.CNN-based dense monocular visual SLAM for real-time UAV exploration in emergency conditions[J].Drones,2022,6(3):79

    • [55] 唐灿,唐亮贵,刘波.图像特征检测与匹配方法研究综述[J].南京信息工程大学学报(自然科学版),2020,12(3):261-273.TANG Can,TANG Lianggui,LIU Bo.A survey of image feature detection and matching methods[J].Journal of Nanjing University of Information Science & Technology(Natural Science Edition),2020,12(3):261-273

    • [56] Li L,Kong X,Zhao XR,et al.Semantic scan context:a novel semantic-based loop-closure method for LiDAR SLAM[J].Autonomous Robots,2022,46(4):535-551

    • [57] Sakkari M,Hamdi M,Elmannai H,et al.Feature extraction-based deep self-organizing map[J].Circuits,Systems,and Signal Processing,2022,41(5):2802-2824

    • [58] Zhu R,Yang M K,Liu W,et al.DeepAVO:efficient pose refining with feature distilling for deep visual odometry[J].Neurocomputing,2022,467:22-35

    • [59] Kim J J Y,Urschler M,Riddle P J,et al.SymbioLCD:ensemble-based loop closure detection using CNN-extracted objects and visual bag-of-words[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).September 27-October 1,2021,Prague,Czech Republic.IEEE,2021:5425

    • [60] Ai Y B,Rui T,Lu M,et al.DDL-SLAM:a robust RGB-D SLAM in dynamic environments combined with deep learning[J].IEEE Access,8:162335-162342

    • [61] Javed Z,Kim G W.PanoVILD:a challenging panoramic vision,inertial and LiDAR dataset for simultaneous localization and mapping[J].The Journal of Supercomputing,2022,78(6):8247-8267

    • [62] Duan R,Feng Y R,Wen C Y.Deep pose graph-matching-based loop closure detection for semantic visual SLAM[J].Sustainability,2022,14(19):11864

    • [63] Memon A R,Wang H S,Hussain A.Loop closure detection using supervised and unsupervised deep neural networks for monocular SLAM systems[J].Robotics and Autonomous Systems,2020,126:103470

    • [64] Li D J,Shi X S,Long Q W,et al.DXSLAM:a robust and efficient visual SLAM system with deep features[C]//2020 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS).October 24,2020-January 24,2021,Las Vegas,NV,USA.IEEE,2020:4958-4965

    • [65] Qin C,Zhang Y Z,Liu Y D,et al.Semantic loop closure detection based on graph matching in multi-objects scenes[J].Journal of Visual Communication and Image Representation,2021,76:103072

    • [66] Godard C,Aodha O M,Brostow G J.Unsupervised monocular depth estimation with left-right consistency[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.July 21-26,2017,Honolulu,HI,USA.IEEE,2017:6602-6611

    • [67] Sang H R,Jiang R,Wang Z P,et al.A novel neural multi-store memory network for autonomous visual navigation in unknown environment[J].IEEE Robotics and Automation Letters,2022,7(2):2039-2046

    • [68] Li G H,Chen S L.Visual slam in dynamic scenes based on object tracking and static points detection[J].Journal of Intelligent & Robotic Systems,2022,104(2):1-10

    • [69] Liu L,Tang T H,Chen J,et al.Real-time 3D reconstruction using point-dependent pose graph optimization framework[J].Machine Vision and Applications,2022,33(2):1-11

    • [70] Xue F,Wang Q,Wang X,et al.Guided feature selection for deep visual odometry[C]//14th Asian Conference on Computer Vision.December 2-6,2018,Perth,Australia.IEEE,2018:293-308

    • [71] Tang Y F,Wei C C,Cheng S L,et al.Stereo visual-inertial odometry using structural lines for localizing indoor wheeled robots[J].Measurement Science and Technology,2022,33(5):055114

    • [72] Bucci A,Zacchini L,Franchi M,et al.Comparison of feature detection and outlier removal strategies in a mono visual odometry algorithm for underwater navigation[J].Applied Ocean Research,2022,118:102961

    • [73] Wu J F,Xiong J,Guo H.Improving robustness of line features for VIO in dynamic scene[J].Measurement Science and Technology,2022,33(6):065204

    • [74] Huang W B,Wan W W,Liu H.Optimization-based online initialization and calibration of monocular visual-inertial odometry considering spatial-temporal constraints[J].Sensors(Basel,Switzerland),2021,21(8):2673

    • [75] Xiang Y,Fox D.DA-RNN:semantic mapping with data associated recurrent neural networks[J].arXiv e-print,2017,arXiv:1703.03098

  • 地址:江苏省南京市宁六路219号    邮编:210044

    联系电话:025-58731025    E-mail:nxdxb@nuist.edu.cn

    南京信息工程大学学报 ® 2024 版权所有  技术支持:北京勤云科技发展有限公司