改进图卷积网络在复杂因素下的步态识别  

Gait recognition based on improved graph convolution network under complex factors

作　　者：吴冬梅[1] 赵梦琦 宋婉莹 王静[1] WU Dong-mei;ZHAO Meng-qi;SONG Wan-ying;WANG Jing(College of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710600,China)

机构地区：[1]西安科技大学通信与信息工程学院,陕西西安710600

出　　处：《计算机工程与设计》2023年第10期3138-3145,共8页Computer Engineering and Design

基　　金：国家自然科学基金-青年基金项目(61901358);陕西省教育厅科研计划基金项目(19JK0528)。

摘　　要：针对人在行走过程中衣着、携带物、视角等因素导致步态识别率显著下降的问题,提出一种加入残差连接和注意力机制的图卷积网络的步态识别方法。以步态骨架序列为输入,采用加入了残差连接的图卷积网络,通过学习骨架数据的时空信息提取出更加精细的步态特征,提高步态特征的表征能力,通过加入注意力机制对各个关节的重要程度进行建模,增大显著区域特征的权重。在数据集CASIA-B中的结果表明,该网络提高了在衣着、携带物等复杂因素下的步态识别率。In view of the significant decline of gait recognition rate caused by people’s clothing,carrying objects and viewing angle during walking,a gait recognition method based on graph convolution network with residual connection and attention mechanism was proposed.The gait skeleton sequence was taken as the input,the graph convolution network with residual connection was used to extract more refined gait features by learning the spatio-temporal information of skeleton data,to improve the representation ability of gait features.The importance of each joint was explicitly modeled by adding attention mechanism to increase the weight of significant regional features.The results in the data set CASIA-B show that the network improves the gait recognition rate under the factors of clothing,carrying objects and so on.

关 键 词：深度学习 生物特征 步态识别 图卷积 注意力机制 残差连接 姿态估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]