CLASSIFICATION OF HUMAN GAIT ACCELERATION DATA USING CONVOLUTIONAL NEURAL NETWORKS
CLASSIFICATION OF HUMAN GAIT ACCELERATION DATA USING CONVOLUTIONAL NEURAL NETWORKS
复制标题
DOI：
10.24507/ijicic.16.02.609
复制
发表时间：
2020-04-01
影响因子：
1
通讯作者：
Zhang, Zhong
中科院分区：
文献类型：
--
作者：
Kreuter, Daniel;Takahashi, Hirotaka;Zhang, Zhong
利用加速度计、陀螺仪等可穿戴传感器进行人体运动分析是普适计算和可穿戴计算中的重要问题之一。灵感来自Akiduki等人的一篇论文。在2018年发布的关于人体步态运动加速度计数据分类的报告中，本文试图使用卷积神经网络对相同的数据进行分类。在2018年的原始论文中，13个记录的测试对象的数据之间存在高度分离，这表明纯粹通过查看运动数据进行分类是可能的。为了使用神经网络进行分类，将给定的时间序列数据转换为三个矩阵(相当于每个像素具有三个通道的图像数据)。使用这些图像作为卷积神经网络的输入，从先前未见的运动数据中分类被摄体编号的准确率达到100%。
The human motion analysis using wearable sensors such as accelerometers and gyroscopes is one of the important issues in ubiquitous and wearable computing. Inspired by a paper by Akiduki et al. that was released in 2018 concerning the classification of human gait motion accelerometer data, this paper attempts to classify that same data using a convolutional neural network. In the original 2018 paper, a high degree of separation was found between the data of the 13 recorded test subjects, suggesting that classification purely by looking at the motion data is possible. For the purpose of classification using the neural network, the given time series data is converted into three matrices (equivalent to image data with three channels per pixel). Using these images as input for a convolutional neural network, an accuracy of 100% was achieved in classifying the subject number from previously unseen motion data.