[1]朱连章,陈殿明,郭加树,等.基于协同LSTM神经网络的人体行为识别研究[J].计算机技术与发展,2018,28(12):79-82.[doi:10．3969/j． issn．1673－629X．2018．12．017]
　ZHU Lianzhang,CHEN Dianming,GUO Jiashu,et al.Research on Human Action Recognition Based on Synergistic LSTM Neural Network[J].,2018,28(12):79-82.[doi:10．3969/j． issn．1673－629X．2018．12．017]
点击复制

基于协同LSTM神经网络的人体行为识别研究()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
28
期数:
2018年12期
页码:
79-82
栏目:
智能、算法、系统工程
出版日期:
2018-12-10

文章信息/Info

Title:
Research on Human Action Recognition Based on Synergistic LSTM Neural Network
文章编号:
1673－629X(2018)12－0079－04
作者:
朱连章; 陈殿明; 郭加树; 张红霞
中国石油大学(华东)计算机与通信工程学院
Author(s):
ZHU Lian－zhang; CHEN Dian－ming; GUO Jia－shu; ZHANG Hong－xia
School of Computer ＆ Communication Engineering，China University of Petroleum，Qingdao 266580，China
关键词:
人体行为识别;  LSTM;  神经网络;  加速度传感器;  陀螺仪
Keywords:
human action recognition; LSTM; neural network; accelerometer; gyroscope
分类号:
TP389.1
DOI:
10．3969/j． issn．1673－629X．2018．12．017
摘要:
在基于传感器的人体行为识别研究中,传统的机器学习方法需要具备一定的人体运动领域知识来做特征提取,而且工程量大。而现有的神经网络模型结构简单,对数据特征的挖掘不充分从而识别准确率不高。针对上述问题,提出一种基于协同LSTM神经网络的人体行为识别方法。该方法首先对LSTM模块的结构进行改进,搭建协同LSTM神经网络;然后使用加速度传感器和陀螺仪获取6轴人体行为数据;再使用滑窗方法和改进的Z-score标准化方法对数据进行预处理;最后利用协同LSTM神经网络、卷积神经网络和LSTM神经网络分别在数据集上进行迭代训练和测试。实验结果表明,基于协同LSTM神经网络的识别模型表现最好,识别准确率为95. 81%,高于CNN的91. 53%和LSTM的90. 47%,证明该方法可以有效地进行人体行为识别。
Abstract:
In the sensor－based human action recognition research，feature extraction by traditional machine learning methods needs lots of work and considerable knowledge in the field of human activity，while the existing neural networks’recognition accuracy is not high for its simple structure and insufficient feature mining． For this，we propose a new human action recognition approach based on synergistic LSTM neural network． Firstly the structure of LSTM module is improved and the synergistic LSTM neural network is built; then six－ax- is human action data are acquired by accelerometer and gyroscope; next，the data will be preprocessed by sliding window and the Z－score normalization; at last，iterative training and testing are performed on the dataset using synergistic LSTM neural network，convolutional neural network and LSTM neural network respectively． The experiment shows that the proposed approach is the best with recognition ac- curacy by 95． 81%，higher than 91． 53% of CNN and 90． 47% of LSTM，which is proved to be effective for human action recognition．

相似文献/References:

[1]宋相法,姚旭.基于多描述子特征编码的人体行为识别[J].计算机技术与发展,2018,28(08):17.[doi:10.3969/ j. issn.1673-629X.2018.08.004]
　SONG Xiang-fa,YAO Xu.Human Activity Recognition Based on Multi-descriptor Feature Coding[J].,2018,28(12):17.[doi:10.3969/ j. issn.1673-629X.2018.08.004]
[2]宋相法,姚旭.基于多特征的深度图像序列人体行为识别[J].计算机技术与发展,2018,28(06):30.[doi:10.3969/ j. issn.1673-629X.2018.06.007]
　SONG Xiang-fa,YAO Xu.Human Activity Recognition from Depth Image Sequences Based on Multiple Features[J].,2018,28(12):30.[doi:10.3969/ j. issn.1673-629X.2018.06.007]
[3]宋相法,吕 明.融合三维骨架和深度图像特征的人体行为识别[J].计算机技术与发展,2019,29(07):55.[doi:10. 3969 / j. issn. 1673-629X. 2019. 07. 011]
　SONG Xiang-fa,LYU Ming.Human Activity Recognition Based on Fusing 3D Skeleton and Depth Image Feature[J].,2019,29(12):55.[doi:10. 3969 / j. issn. 1673-629X. 2019. 07. 011]

更新日期/Last Update: 2018-12-10