摘要
基于信道状态信息的无接触行为识别在很多领域有着重要的应用前景。针对现有方法采用传统的信号统计特征无法同时从低频到高频获得信号的更多细节信号信息,对于信号信息的表征不够明显,识别效果不理想,提出一种基于信道状态信息的无接触人体动作识别方法。首先对采集的CSI数据采用巴特沃斯低通滤波器对数据进行信号预处理;再采用小波包分解算法对CSI信号进行三层小波包分解,同时从第三层提取低频到高频中能表现更多原始信号细节的小波包能量占比特征和小波包系数统计特征组成特征向量;然后再用主成分分析法(PCA)对提取到的特征进行降维处理得到新的特征向量,输入到支持向量机分类器(SVM)并采用十倍交叉验证检验性能。实现了人体日常生活中"蹲下""站起""捡起""跳跃""走路"5种动作的识别,实验结果显示在不同场景下平均识别率可达到95.1%。证明了该方法所提取的特征对信号不同频段细节信息的表征较为明显,能获得较高的识别精度。
The device-independent behavior recognition based on channel state information has important application prospects in many fields. It is impossible for the traditional signal statistical features used in existing methods to obtain more detailed signal information from low frequency to high frequency at the same time,its representation to the signal information is not obvious enough and its recognition effect is unsatisfactory,so a device-independent human motion recognition method based on channel state information is proposed. In this method,a Butterworth low-pass filter is used to perform signal preprocessing of the collected CSI data,and then the wavelet packet decomposition algorithm is adopted to perform three-layer wavelet packet decomposition of CSI signals. Furthermore,the wavelet packet energy ratio feature and wavelet packet coefficient statistical feature that can show more details of the original signals from low frequency to high frequency are extracted from the third layer to form the feature vector. And then,the principal component analysis(PCA)is used to perform dimension reduction for the extracted features,so as to obtain new feature vectors,which are input to the support vector machine(SVM) classifier. The tenfold cross-validation is adopted to test the performance of the method. The test results show that the method can recognize five kinds of actions of "squatting down", "standing up", "picking up", "jumping" and "walking" in daily life of human,and its average recognition rate in different scenarios can reach 95.1%. It has been proved that the features extracted with this method can clearly represent the detailed information at different frequency bands of the signals and obtain higher recognition accuracy.
作者
周祥
常俊
王颖颖
蒙倩霞
王炽
武浩
ZHOU Xiang;CHANG Jun;WANG Yingying;MENG Qianxia;WANG Chi;WU Hao(School of Information Science and Engineering,Yunnan University,Kunming 650091,China)
出处
《现代电子技术》
2021年第17期80-85,共6页
Modern Electronics Technique
基金
国家自然科学基金(61562090)
云南省教育厅科研基金(2019J0007)
云南大学研究生科研创新基金(2019151)。
关键词
行为识别
信道状态信息
小波包分解
主成分分析
信号预处理
十倍交叉验证
behavior recognition
channel state information
wavelet packet decomposition
PCA
signal preprocessing
tenfold cross-validation
