基于残差网络和注意力机制的步态识别算法  

A Gait Recognition Algorithm Based on Residual Network and Attention Mechanism

作　　者：朱磊[1] 仝超 董亮[1] 张聪 ZHU Lei;TONG Chao;DONG Liang;ZHANG Cong(School of Communication and Electronic Engineering,Qiqihar University,Qiqihar 161006,China)

机构地区：[1]齐齐哈尔大学通信与电子工程学院,黑龙江齐齐哈尔161006

出　　处：《电讯技术》2022年第12期1722-1728,共7页Telecommunication Engineering

基　　金：国家自然科学基金资助项目(61501275);中央引导地方科技发展专项项目(SBZY2021E076);黑龙江省自然科学基金项目(联合引导项目)(LH2022F053);黑龙江省博士后科研启动金项目(LBH-Q21195);齐齐哈尔大学学位与研究生教育教学改革研究项目(JGXM_QUG_2019019)。

摘　　要：针对现有行人步态数据集样本较少、多特征融合复杂且识别精准度不高的问题,提出了一种基于残差网络和注意力机制的步态识别算法。首先对每种特征制作标签,然后将表示不同特征的标签做拼接,达到识别多特征的目的。通过迁移ResNet18的预训练模型,并在ResNet18网络第一个和最后一个卷积层后面加入注意力机制,能够加速网络的收敛,提升模型的精准率。在CASIA-B数据集上进行多次实验,结果表明所提出方法能准确识别行人、状态和不同角度三种步态特征,识别的精准率为97.6%。For the problems of few samples,complex multi-feature fusion and low recognition accuracy in the existing pedestrian gait data sets,a gait recognition algorithm based on residual network and attention mechanism is proposed.Firstly,labels are made for each feature,and then the labels representing different features are stitched to achieve the purpose of identifying multiple features.By migrating the pre-training model of ResNet18 and adding an attention mechanism after the first and last convolutional layers of the ResNet18 network,the convergence of the network can be accelerated and the accuracy of the model can be improved.Experiments on the CASIA-B data set show that the proposed method can accurately identify three gait features of pedestrians,states,and different angles,and the recognition accuracy is 97.6%.

关 键 词：步态识别 残差网络 卷积神经网络 机器学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]