第16卷第4期 智能系统学报 Vol.16 No.4 2021年7月 CAAI Transactions on Intelligent Systems Jul.2021 D0:10.11992/tis.202007007 网络出版地址：https:/kns.cnki.net/kcms/detail/23.1538.TP.20210412.1346.010.html 基于二进制生成对抗网络的视觉回环检测研究 杨慧，张婷，金晟，陈良，孙荣川，孙立宁 (苏州大学机电工程学院，江苏苏州215021) 摘要：针对现有的回环检测模型大多基于有监督学习进行训练，需要大量标注数据的问题，提出一种视觉回 环检测新方法，利用生成对抗思想设计一个深度网络，以无监督学习的方式训练该网络并提取高区分度和低维 度的二进制特征。将距离传播损失函数和二值化表示嫡损失函数引入神经网络，将高维特征空间的海明距离 关系传播到低维特征空间并增加低维特征表示的多样性，进而利用B0VW模型将提取的局部特征融合为全局 特征用于回环检测。实验结果表明：相比SFT和OB等特征提取方法，所述方法在具有强烈视角变化和外观 变化的复杂场景下具有更好的性能，可以与AlexNet和AMOSNet等有监督深度网络相媲美。但采用无监督学 习，从根本上避免了费时费力的数据标注过程，特别适用于大规模开放场景的回环检测，同时二进制特征描述 符极大地节约了存储空间和计算资源。 关键词：回环检测；无监督学习；二进制描述符；BoVW;视觉SLAM:生成对抗：特征提取：深度学习 中图分类号：TP181文献标志码：A文章编号：1673-4785(2021)04-0673-10 中文引用格式：杨慧，张婷，金晟，等.基于二进制生成对抗网络的视觉回环检测研究.智能系统学报，2021,16(4)： 673-682. 英文引用格式：YANG Hui,ZHANG Ting,JIN Sheng,.et al.Visual loop closure detection based on binary generative adversarial networkJ.CAAI transactions on intelligent systems,2021,16(4):673-682. Visual loop closure detection based on binary generative adversarial network YANG Hui,ZHANG Ting,JIN Sheng,CHEN Liang,SUN Rongchuan,SUN Lining (School of Mechanical and Electric Engineering,Soochow University,Suzhou 215021,China) Abstract:In view of the problem that the existing loop closure detection models are mostly trained based on supervised learning and require a large amount of labeled data,this paper proposes a new method for visual loop closure detection. The idea of the generative adversarial network is adopted,and thus,a deep neural network is designed and trained through unsupervised learning methods to extract more discriminative binary feature descriptors with low dimensions. The distance propagation loss function and a binarized representation entropy loss function are introduced into the neur- al network.The first loss function can help spread the Hamming distance relationship of the high-dimensional feature space to the low-dimensional feature space,and the second one increases the diversity of the low-dimensional feature representation.The extracted local features are fused into global features by using the BovW model for further loop closure detection.Experimental results show that the proposed method has better performance than feature extraction al- gorithms such as SIFT and ORB in complex scenes that have a strong viewpoint and appearance changes,and its per- formance is comparable with that of supervised deep networks such as AlexNet and AMOSNet.It is especially suitable for loop closure detection in large-scale open scenes because the time-consuming and tedious process of supervised data annotation is completely avoided with the use of unsupervised learning.Moreover,the binary feature descriptors can greatly save storage space and computing resources. Keywords:loop closure detection;unsupervised learning;binary descriptor;BoVW;visual SLAM;generative ad- versarial;feature extraction;deep learning 收稿日期：2020-07-08.网络出版日期：2021-04-12 基金项目：国家自然科学基金面上项目(61673288). 利用三维空间中的信息进行避障、定位以及 通信作者：陈良.E-mail:chenl@suda.edu.cn. 和三维空间中的物体进行交互对于移动机器人等DOI: 10.11992/tis.202007007 网络出版地址: https://kns.cnki.net/kcms/detail/23.1538.TP.20210412.1346.010.html 基于二进制生成对抗网络的视觉回环检测研究 杨慧，张婷，金晟，陈良，孙荣川，孙立宁 （苏州大学 机电工程学院，江苏 苏州 215021） 摘 要：针对现有的回环检测模型大多基于有监督学习进行训练，需要大量标注数据的问题，提出一种视觉回 环检测新方法，利用生成对抗思想设计一个深度网络，以无监督学习的方式训练该网络并提取高区分度和低维 度的二进制特征。将距离传播损失函数和二值化表示熵损失函数引入神经网络，将高维特征空间的海明距离 关系传播到低维特征空间并增加低维特征表示的多样性，进而利用 BoVW 模型将提取的局部特征融合为全局 特征用于回环检测。实验结果表明：相比 SIFT 和 ORB 等特征提取方法，所述方法在具有强烈视角变化和外观 变化的复杂场景下具有更好的性能，可以与 AlexNet 和 AMOSNet 等有监督深度网络相媲美。但采用无监督学 习，从根本上避免了费时费力的数据标注过程，特别适用于大规模开放场景的回环检测，同时二进制特征描述 符极大地节约了存储空间和计算资源。 关键词：回环检测；无监督学习；二进制描述符；BoVW；视觉 SLAM；生成对抗；特征提取；深度学习 中图分类号：TP181 文献标志码：A 文章编号：1673−4785(2021)04−0673−10 中文引用格式：杨慧, 张婷, 金晟, 等. 基于二进制生成对抗网络的视觉回环检测研究 [J]. 智能系统学报, 2021, 16(4): 673–682. 英文引用格式：YANG Hui, ZHANG Ting, JIN Sheng, et al. Visual loop closure detection based on binary generative adversarial network[J]. CAAI transactions on intelligent systems, 2021, 16(4): 673–682. Visual loop closure detection based on binary generative adversarial network YANG Hui，ZHANG Ting，JIN Sheng，CHEN Liang，SUN Rongchuan，SUN Lining (School of Mechanical and Electric Engineering, Soochow University, Suzhou 215021, China) Abstract: In view of the problem that the existing loop closure detection models are mostly trained based on supervised learning and require a large amount of labeled data, this paper proposes a new method for visual loop closure detection. The idea of the generative adversarial network is adopted, and thus, a deep neural network is designed and trained through unsupervised learning methods to extract more discriminative binary feature descriptors with low dimensions. The distance propagation loss function and a binarized representation entropy loss function are introduced into the neur￾al network. The first loss function can help spread the Hamming distance relationship of the high-dimensional feature space to the low-dimensional feature space, and the second one increases the diversity of the low-dimensional feature representation. The extracted local features are fused into global features by using the BoVW model for further loop closure detection. Experimental results show that the proposed method has better performance than feature extraction al￾gorithms such as SIFT and ORB in complex scenes that have a strong viewpoint and appearance changes, and its per￾formance is comparable with that of supervised deep networks such as AlexNet and AMOSNet. It is especially suitable for loop closure detection in large-scale open scenes because the time-consuming and tedious process of supervised data annotation is completely avoided with the use of unsupervised learning. Moreover, the binary feature descriptors can greatly save storage space and computing resources. Keywords: loop closure detection; unsupervised learning; binary descriptor; BoVW; visual SLAM; generative ad￾versarial; feature extraction; deep learning 利用三维空间中的信息进行避障、定位以及 和三维空间中的物体进行交互对于移动机器人等 收稿日期：2020−07−08. 网络出版日期：2021−04−12. 基金项目：国家自然科学基金面上项目 (61673288). 通信作者：陈良. E-mail：chenl@suda.edu.cn. 第 16 卷第 4 期 智 能 系 统 学 报 Vol.16 No.4 2021 年 7 月 CAAI Transactions on Intelligent Systems Jul. 2021