摘要
提出了一种基于无设备的人体行为识别系统,利用Wi-Fi信号的信道状态信息(CSI)来识别3个动态活动:行走、摔倒和坐。该系统只需要一台Wi-Fi路由器作为发射器和一台装有无线网卡的笔记本电脑作为接收器。系统从WiFi信号中提取CSI,然后经过低通滤波以消除噪声,并且为了降低CSI的维度和避免周围噪声所带来的不良影响,在整个CSI数据流中采用了主成分分析(PCA)算法。因此,该系统能够从CSI的时域和频域中得到有用的信号特征值。继而,采用支持向量机(SVM)算法来对人体行为进行分类。为了评优系统的可用性和稳定性,跟5个用户在动态环境中,分别在视距(LOS)和非视距(NLOS)的条件下做了大量的实验,这些实验表明该系统能够得到较高的准确率。
This paper presented a device-free human activity recognition system by using the channel state information (CSI) of Wi-Fi signals, which recognizes three dynamic activities: walk, fall ,and sit. The proposed system required only a Wi-Fi router as a transmitter and a laptop as a receiver. We extracted CSI from Wi-Fi signals and then applied low-pass filter to remove the noise. We applied the Principle Component Analysis (PCA) algorithm across CSI streams to reduce CSI dimensiona[ity and to avoid bad information, which may occurs due to the surrounding noise. Thereafter, we extracted the useful features from time and frequency domain of CSI. Then, we adopted support vector machine to classify the proposed human activity. To examine the feasibility and performance of the proposed system, we implemented hundreds of experiments in Line of sight (LOS) and Non Line of sight (NLOS) seenarios in a dynamic environment with a total of five volunteer users. The experiment results show that our system has gained a high-accuracy rate in both LOS and NLOS.
出处
《武汉理工大学学报》
CAS
北大核心
2016年第4期76-80,共5页
Journal of Wuhan University of Technology
基金
国家自然科学基金面上项目(61373042)和国家自然科学基金青年基金(61502361)