| With the rise of the Internet of things,the use of Wi-Fi Channel State Information(CSI)to realize human behavior perception is of great significance for smart home and smart medical care.Compared with human behavior perception based on sensors and videos,human behavior perception based on Wi-Fi is not affected by environmental factors such as light,does not involve privacy issues,and can be widely deployed at a low cost.In this paper,two CSI based behavioral perception methods are proposed to identify five kinds of daily human behaviors to improve the recognition accuracy.The main research contents are as follows:(1)pretreatment of original CSI data.In this paper,a method combining Butterworth low-pass filtering with local anomaly factor removal is proposed to preprocess the original data and eliminate the noise.In order to avoid excessive computation caused by redundant information,this paper proposes a behavior starting point detection algorithm based on variance to extract effective CSI fragments and accurately judge the beginning and end of behavior actions.(2)a human behavior perception method based on feature extraction and classification was proposed.In this paper,the mean value,standard deviation,mean absolute deviation and third order center distance of the signal are extracted in time domain,and the reliability and robustness of the method are proved by using support vector machine.(3)a human behavior perception model BInd-LSTM based on IndRNN and LSTM was proposed.Traditional method to solve the problems of large calculation and low efficiency,this paper puts forward the BInd-LSTM depth of neural networks,and the sliding window data segmentation,at the same time in the network is a blend of residuals and Batch Normalization(BN),improve the robustness of the model,and from the perspectives of network layer,the contract rate,to explore the factors influencing the accuracy of the model,compared with other models,to prove the validity of the BInd-LSTM model. |
PDF Full Text Request