随着科技的飞速发展和社会的不断进步，公共安全已成为公众关注的焦点。在这样的背景下，生物特征识别技术应用于诸多领域，以确保个人及公共安全。步态识别通过捕捉和分析人们走路的姿态来识别身份，具备远距离性、非接触性和不易伪装的特性，使得它在许多场景中都具有很高的实用价值。然而，步态识别技术作为一个较新的研究方向，仍然面临着诸多挑战，其中，视角、服装和携带背包等变化因素可能导致步态特征的提取和识别变得困难，从而降低步态识别的准确率。为了克服这些挑战，本文提出了基于人体骨架的图卷积网络的步态识别方法，使得步态识别技术能够更好地应对视角、服装和携带背包等变化因素的影响，从而提高步态识别的准确率，论文研究的内容如下：

本文运用卷积神经网络对步态识别进行分析，首先通过对比三种目标检测方法，得出背景减除法可以更好的获取步态目标，其次对二值化图像进行形态学处理，然后将归一化的步态图像送入卷积网络模型中进行训练。

为了更好进行算法的验证，建立融合步态数据集作为训练数据集，探讨了常规的二维卷积网络模型，在模型简单且识别率低的情况下，进一步提高维度，在时间域不变的前提下加入空间域进行特征提取，为了避免网络模型过深，防止梯度消失或梯度爆炸，提出了一种基于三维卷积残差网络的步态识别方法，引入残差学习机制，解决了神经网络在训练过程中出现的梯度消失和模型退化问题，提高了网络的性能，同时设计网络模型的卷积核大小、数目、网络层数及残差模块的布放位置，实验结果表明，提出的三维卷积残差网络在收敛速度及识别率均有提升，但效果依旧不明显。针对上述改进网络模型的识别率不高的问题上，提出了一种基于人体骨架的图卷积网络步态识别模型，通过openpose建立人体步态三维姿态模型，设计图卷积网络模型及网络层的参数，以学习到更稳定且具有辨别度的骨骼特征。

最后，为了验证上述算法的可行性，以Tensorflow为平台进行编程，搭建了一个步态识别系统，并在实际场景中验证了各个算法的效果。实验结果表明，基于人体骨架的图卷积网络模型能分别提取人体骨骼和关节的特征，且可很好的在特征级中融合，进而提取更优的步态特征，提高了识别准确率。

With the rapid development of science and technology and the continuous progress of society, public safety has become the focus of public attention. In this context, biometric recognition technology is applied in many fields to ensure personal and public safety. Gait recognition recognizes identity by capturing and analyzing people's walking posture, and has the characteristics of long distance, non-contact and not easy to camouflage, which makes it have high practical value in many scenes. However, as a relatively new research direction, gait recognition technology still faces many challenges, among which changing factors such as visual angle, clothing and carrying backpack may make it difficult to extract and recognize gait features. So as to reduce the accuracy of gait recognition. In order to overcome these challenges, a gait recognition method based on human skeleton graph convolution network is proposed in this paper, so that the gait recognition technology can better cope with the influence of changing factors such as visual angle, clothing and carrying backpack. in order to improve the accuracy of gait recognition, the contents of this paper are as follows:

In this paper, the convolution neural network is used to analyze gait recognition. Firstly, by comparing three target detection methods, it is concluded that the background subtraction method can better obtain the gait target, and then the binary image is processed morphologically. Then the normalized gait image is sent to the convolution network model for training.

In order to better verify the algorithm, the fused gait data set is established as the training data set, and the conventional two-dimensional convolution network model is discussed. Under the condition of simple model and low recognition rate, the dimension is further improved. Under the premise that the time domain is unchanged, the space domain is added for feature extraction, in order to avoid the network model being too deep and prevent the gradient from disappearing or gradient explosion. In this paper, a gait recognition method based on three-dimensional convolution residual network is proposed, and the residual learning mechanism is introduced to solve the problems of gradient disappearance and model degradation in the training process of the neural network, and the performance of the network is improved. at the same time, the size and number of convolution kernels, the number of network layers and the placement position of residual modules of the network model are designed. The convergence speed and recognition rate of the proposed three-dimensional convolution residual network are improved, but the effect is still not obvious. In order to solve the problem that the recognition rate of the improved network model is not high, a graph convolution network gait recognition model based on human skeleton is proposed. The three-dimensional posture model of human gait is established by openpose, and the parameters of the convolution network model and network layer are designed to learn more stable and distinguishing bone features.

Finally, in order to verify the feasibility of the above algorithms, a gait recognition system is built on the platform of Tensorflow, and the effectiveness of each algorithm is verified in the actual scene. The experimental results show that the graph convolution network model based on human skeleton can extract the features of human bones and joints respectively, and can be well fused in the feature level, so as to extract better gait features and improve the recognition accuracy.

[1]卢世军.生物特征识别技术发展与应用综述[J].计算机安全,2013(1):63-67.

[2]宋磊.王艳杰.数字图像处理中指纹识别技术的分析[J].数字技术与应用,2018,36(12):61-62.

[3]Gui S,Zhou C,Wang H,et al. Application of Voiceprint Recognition Technology Based on Channel Confrontation Training in the Field of Information Security[J]. Electronics,2023,12(15).

[4]Alamgir S,Saiyed U,Kumar R R,et al. Face recognition system with hybrid template protection scheme for Cyber–Physical-Social Services[J]. Pattern Recognition Letters,2023,174.

[5]Huong T T,Du H N,Anh Q T M, et al. A deep learning-based approach for bee sound identification[J]. Ecological Informatics,2023,78:102274.

[6]Niyogi S A,Adelson E H. Analyzing and recognizine walking figures in XYT[R].Proc IEEE Computer Society Conference on Computer Vision and Pattern Recoenition. Texas,USA 1994,64-69.

[7]王科俊,丁欣楠,邢向磊等.多视角步态识别综述[J].自动化学报,2019,45(5):841-852.

[8]Tafazzoli F,Safabakhsh R. Model-based human gait recognition using leg and arm movements[J].Engineering applications of artificial intelligence,2010, 23(8):1237-1246.

[9]Lee L,Grimson W.Gait analysis for recognition and classification[J].Proceedings of the IEEE Conference on Face and Gesture Recognition,2002,155-162.

[10]Liang Y,Li C,Guan Y,et al. Gait recognition based on the golden ratio.[J]. EURASIP J. Image and Video Processing,2016,2016(1):1-8.

[11]王希娟.基于下肢轮廓角度距离和步幅变化的步态识别算法研究[J].计算机应用与软件,2018,35(2):235-241.

[12]Urtasun R.,Fua P.3D tracking for gait characterization and recognition[M].Seoul,Korea:IEEE Computer Society,2004:17-22.

[13]刘勇,李杰,张建林等.基于深度学习的二维人体姿态估计研究进展[J].计算机工程,2021,47(3):16.

[14]许文正,黄天欢,贲晛烨等.跨视角步态识别综述[J].中国图象图形学报,2023,28(5):1265-1286.

[15]Wang G,Wang Y,Zhang H,et al.Exploit the Connectivity:Multi-Object Tracking with TrackletNet[J].2019:482-490.

[16]彭月,甘臣权,张祖凡.人类动作识别的特征提取方法综述[J].计算机应用与软件,2022,39(8):1-14+68.

[17]Tomasz K,Bartosz D,Cíntia F. System for Estimation of Human Anthropometric Parameters Based on Data from Kinect v2 Depth Camera[J]. Sensors (Basel),2023,23(7):3459. .

[18]张秋红,苏锦,杨新锋.基于特征融合和神经网络对步态识别仿真研究[J].计算机仿真,2012,29(8):235-237+245.

[19]李娇娇,孙红岩,董雨等.基于深度学习的3维点云处理综述[J].计算机研究与发展,2022,59(5):1160-1179.

[20]肖雅妮,范馨月,陈文峰.多分支融合局部特征的行人重识别算法[J].计算机工程与应用,2021,57(18):213-219.

[21]李翔,张涛,张哲等.Transformer在计算机视觉领域的研究综述[J].计算机工程与应用, 2023, 59(1):14.

[22]Jia X,Lu H,Yang MH.Visual tracking via adaptive structural local sparse appearance mode[J].IEEE Conference on Computer Vision and Pattern Recognition,2012:1822-1829.

[23]Tang S,Andriluka M,Andres B.Multiple People Tracking by Lifted Multicut and Person Re-identification[J].IEEE Conference on Computer Vision and Pattern Recognition, 2017,3701-3710.

[24]Zeng W,Wang C,Yang F. Silhouette-based gait recognition via deterministic learning[J]. Pattern Recognition,2014,47(11):3568-3584.

[25]Ginting A,Haron C H C,Bencheikh I, et al. Study on characteristics of AlTiN and TiCN coating layers deposited on carbide cutting tools in hard turning of steel: experimental, simulation and optimisation[J]. International Journal of Machining and Machinability of Materials,2021,23(1):88-112.

[26]Marín-Jiménez J M,Castro M F,Carmona-Poyato Á. On how to improve tracklet-based gait recognition systems[J]. Pattern Recognition Letters,2015,68:103-110.

[27]刘语诗,赵秀栩,冯鹤云等.基于拓扑非线性动态建模的神经退行性疾病异常步态识别[J].集成技术,2022,11(4):92-105.

[28]Uddin Z M,Muramatsu D,Takemura N. Spatio-temporal silhouette sequence reconstruction for gait recognition against occlusion[J]. IPSJ Transactions on Computer Vision and Applications,2019,11(1):1-18.

[29]沈雅婷,张炜俊,白郁馨.基于神经网络CNN的三层提取且防过拟合的手势识别算法模型[J].电脑与电信,2023(3):49-54

[30]Yuqi Z,Yongzhen H,Shiqi Y. Cross-view Gait Recognition by Discriminative Feature Learning.[J]. IEEE transactions on image processing,2020,29:1001-1015.

[31]Sakai A,Sogi N,Fukui K. Gait Recognition Based on Constrained Mutual Subspace Method with CNN Features[J].2019 16th International Conference on Machine Vision Applications (MVA),2019,1-6.

[32]黄璇丽,李成明,姜青山.基于深度学习的网络流时空特征自动提取方法[J].集成技术,2020,9(2):60-69.

[33]刘正道,努尔毕亚·亚地卡尔,木特力甫·马木提等.基于优化GaitSet模型的步态识别研究[J].东北师大学报(自然科学版),2022,54(4):77-86.

[34]Christophe C,Valerie R. Walking Direction Estimation Based on Statistical Modeling of HumanGait Features With Handheld MIMU[J]. IEEE/ASME Transactions on Mechatronics,2017,22(6):2502-2511.

[35]李波.无感人脸识别研究综述[J].武汉大学学报(理学版),2023,69(1):115-126.

[36]付莲莲,丁鑫圣,王映龙.基于帧间差分法的口罩佩戴检测算法研究[J].计算机仿真,2022,39(10):250-254.

[37]甄倩倩,张庭亮.步态识别技术研究进展[J].河南教育学院学报(自然科学版),2017,26(2):16-19.

[38]徐中宇,姜洪霖,张忠波.基于新时空融合的步态轮廓分割算法[J].计算机仿真,2010,27(10):202-206.

[39]Passos W,Araujo G,Gois J. A Gait Energy Image-Based System for Brazilian Sign Language Recognition[J]. IEEE Transactions on Circuits and Systems I: Regular Papers,2021,68(11):4761-4771.

[40]钱竞光,宋雅伟,叶强等.步行动作的生物力学原理及其步态分析[J].南京体育学院学报(自然科学版),2006(04):1-7+39.

[41]姚海鹏,王露瑶,刘韵洁. 大数据与人工智能导论[M].人民邮电出版社.2017.

[42]严春满,王铖.卷积神经网络模型发展及应用[J].计算机科学与探索,2021,15(1):27-46.

[43]Xiaoliang W. A class of spectral three-term descent Hestenes-Stiefel conjugate gradient algorithms for large-scale unconstrained optimization and image restoration problems[J]. Applied Numerical Mathematics,2023,192.

[44]Yang L ,Meng T ,Ruijia L.Spike-Based Approximate Backpropagation Algorithm of Brain-Inspired Deep SNN for Sonar Target Classification[J].Computational Intelligence and Neuroscience,2022,1633946-1633946.

[45]Kumar, R. Suresh, Balakrishnan Nagaraj, P. Manimegalai and P. Ajay. Dual feature extraction based convolutional neural network classifier for magnetic resonance imaging tumor detection using U-Net and three-dimensional convolutional neural network[J]. Computers and Electrical Engineering,2022,101:108010.

[46]Vedat A ,M. E E .Densely Connected Dilated Residual Network for Image Denoising: DDR-Net[J].Neural Processing Letters,2022,55(5):5567-5581.

[47]Zhou R Y ,Arun S ,Daniel D.Effects of Image Quality on the Accuracy Human Pose Estimation and Detection of Eye Lid Opening/Closing Using Openpose and DLib[J].Journal of Imaging,2022,8(12):330-330.

[48]Xinyu L ,Na T ,Tao C.Analysis of Gait Characteristics of Patients with Knee Arthritis Based on Human Posture Estimation.[J].BioMed research international,2022,7020804-7020804.

[49]Likai W ,Jinyan C ,Zhenghang C.Multi-stream part-fused graph convolutional networks for skeleton-based gait recognition[J].Connection Science,2022,34(1):652-669.

[50]Zhao B ,Gao X ,Liu J.Spatiotemporal Data Fusion in Graph Convolutional Networks for Traffic Prediction[J].IEEE Access,2020,876632-76641.

[51]Hui L ,F. B S ,Shunlong .Novelty detection of cable-stayed bridges based on cable force correlation exploration using spatiotemporal graph convolutional networks[J].Structural Health Monitoring,2021,20(4):2216-2228.

[52]Ning S ,Ling L ,Jixin Let al.Multi-stream slowFast graph convolutional networks for skeleton-based action recognition[J].Image and Vision Computing,2021:104141-104145.

[53]Xitong Z ,Will R ,Miao Z.Spatiotemporal Graph Convolutional Networks for Earthquake Source Characterization[J].Journal of Geophysical Research: Solid Earth,2022,127(11):e2022JB024401-e2022JB024401.

[54]Wen Y ,Zhang K ,Li Z,et al.A Comprehensive Study on Center Loss for Deep Face Recognition.[J].International Journal of Computer Vision,2019,127(6-7):668-683.

[55]Qilin Z ,Hongmin D ,Kaixuan W .Skeleton Action Recognition Based on Temporal Gated Unit and Adaptive Graph Convolution[J].Electronics,2022,11(18):2973-2973.

[56]Lubin Y ,Lianfang T ,Qiliang D,et al.Multi‐stream adaptive spatial‐temporal attention graph convolutional network for skeleton‐based action recognition[J].IET Computer Vision,2021,16(2):143-158.

[57]Abhilasha Joshi, Neetu Sood, Indu Saini.Analysis of Side View Face Recognition Using Gabor Filter and PCA Techniques[J].ournal of Image Processing & Pattern Recognition Progress,2017,4(2):21-27.

[58]陈海峰.步态识别中的不变特征提取算法设计和研究[D].深圳大学,2017.

[59]Stenum J, Rossi C, Roemmich R T. Two-dimensional video-based analysis of human gait using pose estimation[J]. PLoS Comput Biol,2021,17(4):e1008935.

[60]Tuanjie W ,Rui L ,Huimin Z.Metric-Based Key Frame Extraction for Gait Recognition[J].Electronics,2022,11(24):4177-4177.

[61]Marsico D M ,Mecca A ,Barra S .Walking in a smart city: Investigating the gait stabilization effect for biometric recognition via wearable sensors[J].Computers and Electrical Engineering,2019,80106501-106501.

[62]Singh P J ,Jain S ,Arora S,et al.A Survey of Behavioral Biometric Gait Recognition: Current Success and Future Perspectives[J].Archives of Computational Methods in Engineering,2019,28(1):1-42.

[63]Sien T P J ,Lim H K ,Au P.Deep Learning in Gait Recognition for Drone Surveillance System[J].IOP Conference Series: Materials Science and Engineering,2019,495(1):012031.