首页 > 论文 > 光学学报 > 38卷 > 7期(pp:720001--1)

基于卷积神经网络判别特征学习的行人重识别

Person Re-Identification Based on Convolutional Neural Network Discriminative Feature Learning

  • 摘要
  • 论文信息
  • 参考文献
  • 被引情况
  • PDF全文
分享:

摘要

行人重识别的精确度主要取决于行人的特征描述,基于卷积神经网络的方法具有强大的特征表达能力,可以取得很好的行人重识别效果,但该方法对样本的监督信息利用不充分,且容易过拟合。针对这些问题,提出了一种卷积神经网络判别特征学习的模型,通过加强对样本监督信息的利用和提升对样本间距离关系的约束,使网络得到判别性强的特征。首先,利用批量内样本之间的距离关系来构造难分负样本对,并通过构造的难分负样本对和引入距离门限来改进对比损失函数,充分利用了样本的相似性信息并有效地防止了网络过拟合。其次,通过结合分类模型和验证模型,使批量内样本的所有监督信息都得到了充分利用。在Market1501、DukeMTMC-reID数据库上对所提算法的有效性进行了实验验证,结果表明,所提模型得到的特征具有更强的判别性,而且所提算法的平均识别精确率优于大多数先进算法。

Abstract

The accuracy of person re-identification mainly depends on the feature description of pedestrians. The method based on the convolutional neural network has a powerful feature expression capability and has achieved good results for person re-identification, but its use of the supervised information of the sample is not sufficient and it is easy to overfit. To solve these problems, a convolutional neural network discriminative feature learning model is proposed in this paper. The discriminativity of the network is enhanced by increasing the use of sample supervised information and increasing the constraint on the distance relationship between samples. First of all, we construct the hard-negative sample pairs by using the distance relationship between the samples in batches and improve the contrastive loss function by constructing the hard-negative sample pairs and introducing the distance threshold, making full use of the sample similarity information and effectively preventing the network overfitting. Secondly, by combining the classification model and verification model, all supervised information in the sample in batches is fully utilized. In the experiment, we validate the effectiveness of the proposed algorithm on the Market1501 and DukeMTMC-reID database. The experimental results show that the features of the proposed model obtained in this paper are more discriminative, and the average recognition accuracy of the proposed algorithm is better than that of most advanced algorithms.

Newport宣传-MKS新实验室计划
补充资料

中图分类号:TP391

DOI:10.3788/aos201838.0720001

所属栏目:光计算

基金项目:国家自然科学基金(61472442,61773397,61701524)、陕西省科技新星资助项目(2015kjxx-46)

收稿日期:2018-01-17

修改稿日期:2018-03-13

网络出版日期:--

作者单位    点击查看

陈兵:空军工程大学航空航天工程学院, 陕西 西安 710038
查宇飞:空军工程大学航空航天工程学院, 陕西 西安 710038
李运强:空军工程大学航空航天工程学院, 陕西 西安 710038
张胜杰:空军工程大学航空航天工程学院, 陕西 西安 710038
张园强:空军工程大学航空航天工程学院, 陕西 西安 710038
库涛:空军工程大学航空航天工程学院, 陕西 西安 710038

联系人作者:查宇飞(K13474052583@163.com)

备注:陈兵(1993-),男,博士研究生,主要从事机器视觉及模式识别方面的研究。E-mail: 975571527@qq.com

【1】Sun X W, Xu Q S, Cai Y, et al. Sea sky line detection based on edge phase encoding in complicated background[J]. Acta Optica Sinica, 2017, 37(11): 1110002.
孙熊伟, 徐青山, 蔡熠, 等. 基于边缘相位编码的复杂背景下海天线检测[J]. 光学学报, 2017, 37(11): 1110002.

【2】Wang B, Su Y M, Wan L, et al. Sea sky line detection method of unmanned surface vehicle based on gradient saliency[J]. Acta Optica Sinica, 2016, 36(5): 0511002.
王博, 苏玉民, 万磊, 等. 基于梯度显著性的水面无人艇的海天线检测方法[J]. 光学学报, 2016, 36(5): 0511002.

【3】Xu L, Zhao H T, Sun S Y. Monocular infrared image depth estimation based on deep convolutional neural networks[J]. Acta Optica Sinica, 2016, 36(7): 0715002.
许路, 赵海涛, 孙韶媛. 基于深层卷积神经网络的单目红外图像深度估计[J]. 光学学报, 2016, 36(7): 0715002.

【4】Zheng Z, Zheng L, Yang Y. Adiscriminatively learned CNN embedding for person re-identification[J]. ACM Transactions on Multimedia Computing Communications and Applications, 2017, 14(1): 13.

【5】Yi D, Lei Z, Liao S, et al. Deep metric learning for person re-identification[C]. Proceedings of IEEE, 2014: 34-39.

【6】Zhong Z, Zheng L, Cao D, et al. Re-ranking person re-identification with k-reciprocal encoding[C]. Proceedings of IEEE, 2017: 3652-3661.

【7】Wang F, Zuo W, Lin L, et al. Joint learning of single-image and cross-image representations for person re-identification[C]. Proceedings of IEEE, 2016: 1288-1296.

【8】Varior R R, Shuai B, Lu J, et al. A siamese long short-term memory architecture for human re-identification[C]. Proceedings of ECCV, 2016, 9911: 135-153.

【9】Li W, Zhao R, Xiao T, et al.DeepReID: deep filter pairing neural network for person re-identification[C]. Proceedings of IEEE, 2014: 152-159.

【10】Cheng D, Gong Y, Zhou S, et al. Person re-identification by multi-channel parts-based CNN with improved triplet loss function[C]. Proceedings of IEEE, 2016: 1335-1344.

【11】Zheng Z, Zheng L, Yang Y. Unlabeled samples generated by GAN improve the person re-identification baseline in vitro[C]. Proceedings of IEEE, 2017: 3774-3782.

【12】Sun Y, Wang X, Tang X. Deep learning face representation by joint identification-verification[C]. Proceedings of NIPS, 2014: 1988-1996.

【13】Liu J, Zha Z J, Tian Q I, et al. Multi-scale triplet CNN for person re-identification[C]. Proceedings of ACM, 2016: 192-196.

【14】Zheng L, Shen L, Tian L, et al. Scalable person re-identification: a benchmark[C]. Proceedings of IEEE, 2015: 1116-1124.

【15】Varior R R, Haloi M, Wang G, et al. Gated siamese convolutional neural network architecture for human re-identification[C]. Proceedings of ECCV, 2016, 9912: 791-808.

【16】Liao S, Hu Y, Zhu X, et al. Person re-identification by local maximal occurrence representation and metric learning[C]. Proceedings of the IEEE, 2015: 2197-2206.

引用该论文

Chen Bing,Zha Yufei,Li Yunqiang,Zhang Shengjie,Zhang Yuanqiang,Ku Tao. Person Re-Identification Based on Convolutional Neural Network Discriminative Feature Learning[J]. Acta Optica Sinica, 2018, 38(7): 0720001

陈兵,查宇飞,李运强,张胜杰,张园强,库涛. 基于卷积神经网络判别特征学习的行人重识别[J]. 光学学报, 2018, 38(7): 0720001

您的浏览器不支持PDF插件,请使用最新的(Chrome/Fire Fox等)浏览器.或者您还可以点击此处下载该论文PDF