基于穿墙信道状态信息的行为识别方法  

Behavior Recognition Method Based on Channel State Information in Through-the-Wall

作　　者：蒙倩霞 余江[1] 常俊[1] 浦钰 陈澄 MENG Qianxia;YU Jiang;CHANG Jun;PU Yu;CHEN Cheng(School of Information,Yunnan University,Yunnan,Kunming 650500,China)

机构地区：[1]云南大学信息学院,昆明650500

出　　处：《计算机工程》2021年第8期201-209,共9页Computer Engineering

基　　金：国家自然科学基金(61162004);云南省教育厅科学研究基金(2019J0007);云南大学研究生科研创新基金(2019151)。

摘　　要：Wi-Fi CSI提供的被动式行为识别方法在许多场景得到了应用,但现有系统较少考虑信号穿墙的场景,导致信号穿墙后识别精度急剧下降。为提高系统的适应性,对信号穿墙后的行为识别方法进行研究,提出一种基于信道状态信息(CSI)的穿墙行为识别方法。根据CSI数据变化的特性,在预处理阶段,对CSI数据进行相位校准来证明原始CSI矩阵具有低秩性,并对其进行低秩矩阵分解,消除无用的静态CSI分量,凸显信号穿过墙壁后被掩盖的动态CSI分量。在识别行为阶段,利用时间反演算法解决CSI数据维度过高的问题,并简化计算。实验结果表明,与传统行为识别方法相比,该方法可大幅提升穿墙场景下的行为识别精度,在室内视距、室内非视距、穿墙的场景下平均识别精度分别可达94.1%、92.3%、90.7%。The passive behavior recognition method provided by Wi-Fi CSI has been applied in many scenarios,but the existing systems hardly consider the scene where the signal passes through the wall,or the sharp decline in recognition accuracy after the signal passes through the wall.In order to improve the adaptability of the system,the method of recognizing the behavior of the signal passing through the wall is studied,and a recognition method based on Channel State Information(CSI)is proposed.According to the changing characteristics of the CSI data,in the preprocessing stage,the phase calibration of the CSI data is carried out,and then the low rank of the original CSI matrix is proved and decomposed by the low rank matrix to eliminate the useless static CSI components,so that the dynamic CSI components which are masked to a large extent after the signal passes through the wall are highlighted.In the recognition behavior stage,the time reversal algorithm is used to solve the problem that the dimension of CSI data is too high,and the calculation is simplified.Experimental results show that compared with traditional behavior recognition methods,the accuracy of behavior recognition in through-wall scene can be greatly improved.The average recognition accuracy can reach 94.1%,92.3%and 90.7%respectively within the range of indoor visibility,beyond the range of indoor visibility and in through-wall scenes.

关 键 词：信道状态信息 行为识别 相位校准 低秩矩阵分解 静态分量消除 时间反演算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]