基于CLSTM的步态分类方法

步态分类在人体运动能量消耗评估等应用中具有重要意义，提高分类精度和降低对统计特征的依赖是步态分类的研究热点。采用传统的步态分类方法提取的步态特征用于细分化步态时不能得到较好的效果。考虑到步态的连续性和不同轴之间信号的相关性，本文提出了基于CLSTM的步态分类方法：采用卷积神经网络（CNN）操作，通过计算多轴步态数据提取步态特征；基于长短期记忆（LSTM）构建步态时间序列模型，学习步态特征图时间维度上的长期依赖性。基于USC-HAD数据集的实验结果表明，用此方法提取了步态序列特征，很好地利用了步态时间序列特点，提升了11种步态的分类精度。

A Gait Pattern Classification Method Based on CLSTM

Gait classification is an effective method for the assessment of human motion energy consumption,in which the key issue is to improve its classification accuracy and decrease the dependence on statistic features.Aiming at the shortcoming of the traditional gait classification methods in classifying subdivided gaits,this paper proposes a CLSTM method by considering the continuity of the gait and the signal correlation among different axes.By means of CNN convolution operation,this proposed method can extract the gait features by calculating the gait data among multi-axis.Besides,the present method utilizes the LSTM-based gait time series model to learn long-term dependent relation on gait features in the time dimension.Finally,it is illustrated via experiment on USC-HAR datasets that the proposed method can extract gait sequence features and effectively utilize the characteristics of gait time-series to raise classification accuracy in 11 gaits pattern.