摘要
当前,多数步态识别方法关注于步态序列单一时间尺度建模,忽略了不同时间尺度的信息交互。基于此,提出了一种双尺度时间特征表示网络。该方法聚合两个时间尺度特征来获取步态的运动表示,并将两个时间尺度上特征进行融合,实现信息交互。通过多视角识别实验验证,该方法在数据集CASIA-B上的性能超越了主流的步态识别方法,在NM、BG和CL条件下Rank-1准确率分别达到97.8%、93.1%以及80.6%。
At present,most gait recognition methods focus on the modeling of a single time scale of gait sequences,ignoring the information interaction of different time scales.Based on this,a dual-scale temporal feature representation network is proposed.This method aggregates two time level features to obtain the motion representation of gait,and fuses the features on the two time scales to achieve information interaction.Through experimental verification,the performance of this method on the data set CASIA-B sur-passes the mainstream gait recognition method,and the Rank-1 accuracy rate reaches 97.8%,93.1%and 80.6%under NM,BG and CL conditions,respectively.
作者
魏永超
徐未其
朱泓超
朱姿翰
刘伟杰
Wei Yongchao;Xu Weiqi;Zhu Hongchao;Zhu Zihan;Liu Weijie(Scientific Research Office,China Civil Aviation Flight Academy,Deyang 618307,China;School of Civil Aviation Safety Engineering,China Civil Aviation Flight Academy,Deyang 618307,China;School of Avionics and Electrical Engineering,China Civil Aviation Flight Academy,Deyang 618307,China)
出处
《现代计算机》
2024年第6期8-13,55,共7页
Modern Computer
基金
西藏科技厅重点研发计划(XZ202101ZY0017G)
四川省科技厅重点研发项目(2022YFG0356)
中国民用航空飞行学院科研基金(J2020-040、CJ2020-01)。
关键词
步态识别
时间尺度
空间特征
多视角识别
gait recognition
time level
spatial features
multi-view recognition
