Abstract In visual loop closure detection under strong scene changes, the feature descriptors extracted by the existing deep learning methods cannot be distinguished well. Aiming at this problem, the multi-constraint distance relationship is analyzed, and a multi-constraint deep distance learning method for visual loop closure detection is proposed. Firstly, the original images are mapped to feature descriptors by any convolutional neural network in the low-dimensional feature space. Then, a multi-constraint loss function is proposed to constrain the distance relationships among feature descriptors, and a multi-constraint training sample set is automatically constructed online to extract more discriminative low-dimensional feature descriptors. Experiments on New College and TUM datasets show that the proposed method improves the performance of loop closure detection under strong scene changes.
Fund:Supported by National Natural Science Foundation of China(No.61673288)
About author:: (CHEN Liang(Corresponding author), Ph.D., associate professor. His research interests include deep learning.);(JIN Sheng, master student. His research interests include visual loop closure detection.);(YANG Hui, master student. Her research interests include visual loop closure detection.);(GAO Yu, Ph.D., lecturer. His research interests include robotic visual perception.);(SUN Rongchuan, Ph.D., associate profe-ssor. His research interests include visual SLAM.);(SUN Lining, Ph.D., professor. His research interests include advanced robotic technologies.)
[1] 梁志伟,陈燕燕,朱松豪,等.基于视觉词典的单目视觉闭环检测算法.模式识别与人工智能, 2013, 26(6): 561-570. (LIANG Z W, CHEN Y Y, ZHU S H, et al. Loop Closure Detection Algorithm Based on Monocular Vision Using Visual Dictionary. Pattern Recognition and Artificial Intelligence, 2013, 26(6): 561-570.) [2] 刘 强,段富海,桑 勇,等.复杂环境下视觉SLAM闭环检测方法综述.机器人, 2019, 41(1): 112-123, 136. (LIU Q, DUAN F H, SANG Y, et al. A Survey of Loop-Closure Detection Method of Visual SLAM in Complex Environments. Robot, 2019, 41(1): 112-123, 136.) [3] 王小龙,彭国华.基于Rényi散度最大化的多特征闭环检测.模式识别与人工智能, 2018, 31(5): 389-397. (WANG X L, PENG G H. Loop Closure Detection Based on Maximizing Rényi Divergence of Multiple Features. Pattern Recognition and Artificial Intelligence, 2018, 31(5): 389-397.) [4] LOWE D G. Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision, 2004, 60(2): 91-110. [5] BAY H, ESS A, TUYTELAARS T, et al. Speeded-up Robust Features(SURF). Computer Vision and Image Understanding, 2008, 110(3): 346-359. [6] RUBLEE E, RABAUD V, KONOLIGE K, et al. ORB: An Efficient Alternative to SIFT or SURF // Proc of the International Conference on Computer Vision. Washington, USA: IEEE, 2011: 2564-2571. [7] ROSTEN E, DRUMMOND T. Machine Learning for High-Speed Corner Detection // Proc of the European Conference on Computer Vision. Berlin, Germany: Springer, 2006: 430-443. [8] CALONDER M, LEPETIT V, STRECHA C, et al. BRIEF: Binary Robust Independent Elementary Features // Proc of the European Conference on Computer Vision. Berlin, Germany: Springer, 2010:778-792. [9] CHEN Z, LAM O, JACOBSON A,et al. Convolutional Neural Network-Based Place Recognition [C/OL]. [2020-02-02]. https://arxiv.org/ftp/arxiv/papers/1411/1411.1509.pdf. [10] CHEN Z T, JACOBSON A, SUNDERHAUF N, et al. Deep Lear-ning Features at Scale for Visual Place Recognition // Proc of the IEEE International Conference on Robotics and Automation. Wa-shington, USA: IEEE, 2017: 3223-3230. [11] SUNDERHAUF N, SHIRAZI S, DAYOUB F, et al. On the Performance of ConvNet Features for Place Recognition // Proc of the IEEE/RSJ International Conference on Intelligent Robots and Systems. Washington, USA: IEEE, 2015: 4297-4304. [12] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet Cla-ssification with Deep Convolutional Neural Networks // BARTLETT P, PEREIRA F, BURGES C J C, et al., eds. Advances in Neural Information Processing Systems 25. Cambridge, USA: The MIT Press, 2012: 1-9. [13] 赵 洋,刘国良,田国会,等.基于深度学习的视觉SLAM综述.机器人, 2017, 39(6): 889-896. (ZHAO Y, LIU G L, TIAN G H, et al. A Survey of Visual SLAM Based on Deep Learning. Robot, 2017, 39(6): 889-896.) [14] DING S Y, LIN L, WANG G R, et al. Deep Feature Learning with Relative Distance Comparison for Person Re-identification. Pattern Recognition, 2015, 48(10): 2993-3003. [15] SCHROFF F, KALENICHENKO D, PHILBIN J. FaceNet: A Unified Embedding for Face Recognition and Clustering // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Washington, USA: IEEE, 2015: 815-823. [16] SIMONYAN K, ZISSERMAN A. Very Deep Convolutional Networks for Large-Scale Image Recognition[C/OL]. [2020-02-02]. https://arxiv.org/abs/1409.1556.pdf. [17] ZHENG L, YANG Y, TIAN Q. SIFT Meets CNN: A Decade Survey of Instance Retrieval. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(5): 1224-1244. [18] AZIZPOUR H, RAZAVIAN A S, SULLIVAN J, et al. Factors of Transferability for a Generic ConvNet Representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(9): 1790-1802. [19] CUMMINS M J, NEWMAN P. FAB-MAP: Probabilistic Localization and Mapping in the Space of Appearance. International Journal of Robotics Research, 2008, 27(6): 647-665. [20] STURM J, ENGELHARD N, ENDRES F, et al. A Benchmark for the Evaluation of RGB-D SLAM Systems // Proc of the IEEE/RSJ International Conference on Intelligent Robots and Systems. Wa-shington, USA: IEEE, 2012: 573-580. [21] GAO X, ZHANG T. Unsupervised Learning to Detect Loops Using Deep Neural Networks for Visual SLAM System. Autonomous Robots, 2017, 41(1): 1-18.
2020, Vol. 33 
