CGAC:一种基于CSI的人体动作识别方法
CGAC:a CSI-based human activity recognition method
苏健 1郑毓煌 1陈思光2
作者信息
- 1. 南京信息工程大学软件学院,江苏南京 210044
- 2. 南京邮电大学物联网学院,江苏南京 210003
- 折叠
摘要
WiFi的信道状态信息(CSI)在人体动作识别(HAR)领域具有广泛的应用前景.目前基于CSI的HAR大多在准确率以及不同环境中的鲁棒性上存在不足.针对这类问题,提出了一种结合卷积神经网络、门控循环单元以及注意力机制的复合人体动作识别模型(CGAC).首先使用CNN对输入数据进行时序特征提取,通过池化操作减小特征尺寸,再使用BiGRU对时序特征进行建模,通过注意力机制增强对关键特征的关注度.在3个公开数据集进行实验,CGAC在UT-HAR数据集中达到了 99.70％的准确率,在NTU-Fi的HAR数据集中达到了 97.50％的准确率,在Human-ID数据集上达到了 97.81％的准确率,实验结果表明CGAC模型高于该领域现有方法的准确率,证明了方法的有效性.
Abstract
Channel state information(CSI)of WiFi has a wide range of applications in the field of human action recognition(HAR).Most methods of CSI-based HAR are deficient in accuracy and lack robustness in different environments.To address these issues,this paper proposes a composite human action recognition model(CGAC)that combines convolutional neural networks(CNNs),gated recurrent units,and attention mechanisms.First,temporal features are extracted from the input data using CNNs.Second,the feature size is reduced by the pooling operation.Third,the temporal features are modeled by using BiGRU.Thus,the attention to the key features is enhanced by the attention mechanism.Experiments are conducted on three public datasets,and the results show that CGAC obtains a higher accuracy than that of any other existing methods:99.70％accuracy on the UT-HAR dataset,97.50％on the HAR dataset of NTU-Fi,and 97.81％on the Human-ID dataset,validating its effectiveness.
关键词
人体动作识别/信道状态信息/深度学习/卷积神经网络/门控循环单元/注意力机制Key words
human activity recognition(HAR)/channel state information(CSI)/deep learning/convolutional neural network(CNN)/gate recurrent unit(GRU)/attention mechanism引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据