一种时空卷积的步态识别方法

doi:10.19665/j.issn1001-2400.2021.04.019

西安电子科技大学学报 ›› 2021, Vol. 48 ›› Issue (4): 144-150.doi: 10.19665/j.issn1001-2400.2021.04.019

• 计算机科学与技术&网络空间安全 • 上一篇下一篇

一种时空卷积的步态识别方法

许缓缓^1,²(),李洪梅^1,²(),李富余³(),孙学梅^1,²()

1.天津市自主智能技术与系统重点实验室,天津 300387
2.天津工业大学计算机科学与技术学院,天津 300387
3.花旗金融信息服务(中国)有限公司,上海 201203

收稿日期:2020-05-15 出版日期:2021-08-30 发布日期:2021-08-31
通讯作者: 孙学梅
作者简介:许缓缓(1987—),女,天津工业大学硕士研究生,E-mail: xuhh@tiangong.edu.cn|李洪梅(1998—),女,天津工业大学硕士研究生,E-mail: 15122973198@163.com|李富余(1989—),女,硕士,E-mail: lifuyu@csu.edu.cn
基金资助:
天津市自然科学基金(19JCYBJC15400);天津市科技计划项目(18JCTPJC61900);天津市科技计划项目(20YDTPJC01200)

Gait recognition method based on spatial-temporal convolution

XU Huanhuan^1,²(),LI Hongmei^1,²(),LI Fuyu³(),SUN Xuemei^1,²()

1. Tianjin Key Laboratory of Autonomous Intelligence Technology and Systems,Tianjin 300387,China
2. School of Computer Science and Technology,Tiangong University,Tianjin 300387,China
3. Citigroup Services and Technology(China)Limited,Shanghai 201203,China

Received:2020-05-15 Online:2021-08-30 Published:2021-08-31
Contact: Xuemei SUN

摘要/Abstract

摘要：

由于步态识别的准确率会受到外部因素的影响,例如拍摄角度、行人穿着以及携带包的状态变化等,所以提出基于时空卷积的步态识别算法。该方法通过卷积神经网络提取步态特征,并通过对相邻帧进行重复提取来弥补当前帧丢失的信息,从而在步态轮廓序列图里挖掘更丰富的时空信息。最后,在CASIA-B数据集上对所提出的方法进行验证。实验结果表明,所提出的方法在正常行走、携带包和穿外套的情况下,Rank-1准确度都得到了很大的提升。

关键词: 步态识别, 时空卷积, 步态特征, 时空信息

Abstract:

As a unique biological characteristic that can be recognized from a distance,gait has a wide application prospect in the fields of identity identification,public security and medical diagnosis.However,the accuracy of gait recognition will be affected by external factors,such as the shooting angle,the pedestrian’s wearing and the change in the status of carrying a bag.Based on the above problems,this paper puts forward the gait recognition method based on spatial-temporal convolution,which uses the convolution neural network to extract gait features,and employs the repeated extraction of the gait features of adjacent frames to make up for the missing information,so that more spatial-temporal information can be obtained.Finally,the proposed method is validated on the CASIA-B dataset.Experimental results show that the proposed method can improve the rank-1 accuracy when a pedestrian walks normally,carrying a bag and wearing a coat.

Key words: gait recognition, spatial-temporal convolution, gait feature, spatial-temporal information

中图分类号:

TP391.41

许缓缓,李洪梅,李富余,孙学梅. 一种时空卷积的步态识别方法[J]. 西安电子科技大学学报, 2021, 48(4): 144-150.

XU Huanhuan,LI Hongmei,LI Fuyu,SUN Xuemei. Gait recognition method based on spatial-temporal convolution[J]. Journal of Xidian University, 2021, 48(4): 144-150.

图/表 7

图1

图2

图3

图4

表1

表2

表3

参考文献 19

[1]	HAN J, BHANU B. Individual Recognition Using Gait Energy Image[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28(2):316-322. doi: 10.1109/TPAMI.34
[2]	CHAO H, HE Y, ZHANG J, et al. GaitSet:Regarding Gait as a Set for Cross-View Gait Recognition[C]. Proceedings of the Thirty-Third AAAI Conference on Artifical Intelligence (AAAI) 2019:8126-8133.
[3]	ZHANG Z, TRAN L, YIN X, et al. Gait Recognition Via Disentangled Representation Learning[C]//Proceedings of the 2019 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Long Beach:CA,USA, 2019:4705-4714.
[4]	ZHANG K, LUO W, MA L, et al. Learning Joint Gait Representation via Quintuplet Loss Minimization[C]//Proceedings of the 2019 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Long Beach:CA,USA, 2019:4695-4704.
[5]	JI S, XU W, YANG M, et al. 3D Convolutional Neural Networks for Human Action Recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(1):221-231. doi: 10.1109/TPAMI.2012.59
[6]	KARPATHY A, TODERICI G, SHETTY S, et al. Large-Scale Video Classification with Convolutional Neural Networks[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Columbus:OH,USA, 2014:1725-1732.
[7]	NIXON M S, CARTER J N, NASH J M, et al. Motion Analysis and Tracking[C]//Proceedings of the IEEE Colloquium on Motion Analysis and Tracking.London:UK, 1999:5-6.
[8]	陈昌红; 赵恒; 梁继民; 焦李成. 特征选择对FHMM性能影响研究[J]. 西安电子科技大学学报, 2010, 37(5):934-940.
	CHEN Changhong, ZHAO Heng, LIANG Jimin, JIAO Licheng. Influence of Feature Selection on FHMM[J]. Journal of Xidian University, 2010, 37(5):934-940.
[9]	BASHIR K, XIANG T, GONG S. Gait Recognition Using Gait Entropy Image[C]//Proceedings of the 3rd International Conference on Imaging for Crime Detection and Prevention(ICDP).London:IET, 2009:1-6.
[10]	毛莺池, 张建华, 陈豪. 一种多视图深度融合的连续性缺失补全方法[J]. 西安电子科技大学学报, 2019, 46(2):61-68.
	MAO Yingchi, ZHANG Jianhua, CHEN Hao. Successive Missing Completion Based on Deep Fusion from Multiple Views[J]. Journal of Xidian University, 2019, 46(2):61-68.
[11]	YIWEIH, JUNPING Z, HONGMING S, et al. Multi-task GANs for View-Specific Feature Learning in Gait Recognition[J]. IEEE Transactions on Information Forensics and Security, 2019, 14(1):102-113. doi: 10.1109/TIFS.2018.2844819
[12]	DAS D, AGARWAL A, CHATTOPADHYAY P, et al. RGait-NET:An Effective Network for Recovering Missing Information from Occluded Gait Cycles[J]. arxiv:1912.067565, 2019.
[13]	LI W, ZHU X, GONG S. Harmonious Attention Network for Person Re-Identification[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City:UT,USA, 2018:2285-2294.
[14]	FU Y, WEI Y, ZHOU Y, et al. Horizontal Pyramid Matching for Person Re-Identification[C]//Proceedings of the Thirty-First Innovative Applications of Artificial Intelligence Conference(AAAI), 2019:8295-8302.
[15]	TAKEMURA N, MAKIHARA Y, MURAMATSU D, et al. Multi-View Large Population Gait Dataset and Its Performance Evaluation for Cross-View Gait Recognition[J]. IPSJ Transactions on Computer Vision and Applications, 2018, 10(1):4. doi: 10.1186/s41074-018-0039-6
[16]	HU M, WANG Y, ZHANG Z, et al. View-Invariant Discriminative Projection for Multi-View Gait-Based Human Identification[J]. IEEE Transactions on Information Forensics and Security, 2013, 8(12):2034-2045. doi: 10.1109/TIFS.2013.2287605
[17]	KUSAKUNNIRAN W, WU Q, ZHANG J, et al. Recognizing Gaits Across Views Through Correlated Motion Co-Clustering[J]. IEEE Transactions on Image Processing, 2014, 23(2):696-709. doi: 10.1109/TIP.2013.2294552
[18]	WU Z, HUANG Y, WANG L, et al. A Comprehensive Study on Cross-View Gait Based Human Identification with Deep CNNs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence(TPAMI), 2017, 39(2):209-226.
[19]	YU S, TAN D, TAN T. A Framework for Evaluating the Effect of View Angle,Clothing and Carrying Condition on Gait Recognition[C]//Proceedings of the 18th International Conference on Pattern Recognition(ICPR).Hong Kong:China, 2006:441-444.

Gallery NM#1-4		0°~180°											平均值
Probe		0°	18°	36°	54°	72°	90°	108°	126°	144°	162°	180°	平均值
正常	ViDP^[16]				59.1		50.2		57.5
	CMCC^[17]	46.3			52.4		48.3		56.9
	CNN-LB^[18]	54.8			77.8		64.9		76.1
	GaitSet^[2]	64.6	83.3	90.4	86.5	80.2	75.5	80.3	86	87.1	81.4	59.6	79.5
	文中	69.8	82.2	88.5	86.7	79.5	74.0	79.6	86.0	87.6	81.5	65.2	80.0
携带包	GaitSet^[2]	55.8	70.5	76.9	75.5	69.7	63.4	68	75.8	76.2	70.7	52.5	68.6
	文中	59.6	72.0	78.2	75.8	69.8	64.6	68.2	77.0	76.5	70.8	56.3	69.9
穿外套	GaitSet^[2]	29.4	43.1	49.5	48.7	42.3	40.3	44.9	47.4	43	35.7	25.6	40.9
	文中	31.8	46.5	51.0	48.2	44.0	41.7	43.2	45.3	40.9	35.2	27.5	41.4

Gallery NM#1-4	0°~180°				平均值	36°~144°			平均值
Probe NM #5-6	0°	54°	90°	126°	平均值	54°	90°	126°	平均值
CMCC^[17]	46.3	52.4	48.3	56.9	51.0
ViDP^[16]	59.1	50.2	57.5		83.5	76.7	80.7	80.3
MT^[18]	54.8	77.8	64.9	76.1	68.4	90.8	85.8	90.4	89.0
文中	69.8	86.7	74.0	86.0	79.1	96.5	93.3	96.1	95.3

Gallery NM#1-4	0°~180°				平均值	36°~144°			平均值
Probe NM #5-6	0°	54°	90°	126°	平均值	54°	90°	126°	平均值
ViDP^[16]		64.2	60.4	65.0		87	87.7	89.3	88.0
LB^[18]	82.6	94.3	87.4	94.0	89.6	98.0	98.0	99.2	98.4
GaitNet^[3]	91.2	95.6	92.6	96.0	93.9	99.1	99.0	99.2	99.1
文中	91.3	96.7	90.3	97.5	94.0	99.7	97.2	99.8	98.9
Probe BG #1-2	0°	54°	90°	126°	平均值	54°	90°	126°	平均值
LB-subGEI^[18]	64.2	76.9	63.1	76.9	70.3	89.2	84.3	91.0	88.2
GaitNet^[3]	83.0	86.6	74.8	85.8	82.6	90.0	85.6	92.7	89.4
文中	81.8	89.7	78.7	90.9	85.3	95.0	92.0	96.8	94.6
Probe CL #1-2	0°	54°	90°	126°	平均值	54°	90°	126°	平均值
LB-subGEI^[18]	37.7	61.1	54.6	59.1	53.1	77.3	74.5	74.5	75.4
GaitNet^[3]	42.1	70.7	70.6	69.4	63.2	80.0	81.2	79.4	80.2
文中	60.8	74.1	64.3	71.3	67.6	82.3	79.0	78.5	79.9

一种时空卷积的步态识别方法

Gait recognition method based on spatial-temporal convolution

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 19

相关文章 3

Metrics

本文评价

推荐阅读 10

[1]	戚艳军,孔月萍,王佳婧,朱旭东. 一种LSTM与CNN相结合的步态识别方法[J]. 西安电子科技大学学报, 2021, 48(5): 78-85.
[2]	陈昌红;赵恒;梁继民;焦李成. 特征选择对FHMM性能影响研究[J]. J4, 2010, 37(5): 934-940.
[3]	陈实1;田玉敏1;黄万红2;高有行1. 用时变模板自动识别行人的步态 [J]. J4, 2007, 34(4): 605-610.