Research On Human Motion Detection Technology Based On Wireless Channel State Information

Nowadays,human activity detection has become a key research direction in the field of medical health,security and smart home,attracting numerous researcher interests.Traditional sensor-based and computer vision-based activity detection technologies have such problems as expensive equipment,difficult deployment and privacy.With the popularity of Wi-Fi network,many researchers have focused on activity detection based on Wi-Fi network.The researchers first used the received signal strength indicator(RSSI)to detect the activity,but the accuracy of this method was limited.In recent years,researchers have found that fine-grained channel state information(CSI)can provide subcarrier-level data for detecting activities.In this method,different activities have different effects on CSI.By analyzing different CSI,the different activities can be identified.Specifically,the main work of this paper are as follows:(1)The amplitude of Wi-Fi Channel State Information(CSI)is used to distinguish various standing and sitting activities.Analyzing the CSI amplitude waveforms,the important phenomena including strong influence of indoor environment and activity-induced waveform symmetry are found.The findings enlighten the design of this system.To preprocess data,firstly the outliers and noises are filtered out from raw CSI amplitude data,then a two-tier sliding window thresholded method and adjacent-activity combination method are proposed to accurately segment different activities,succeeded by data resampling.Subsequently,the processed CSI amplitude data in different scenarios are input into Convolution Neural Network(CNN)which is trained to obtain a suitable model for online activity recognition.Finally,a detection method is proposed to improve the accuracy of activity recognition.The method is fulfilled based on the confidence of CNN outputs,the mutual restriction between activities and the activity-waveform symmetry.The performance of the system is verified in different scenarios.Comprehensive experiments demonstrate the robustness and high accuracy of this system.(2)This paper also proposes a sleep monitoring system based on channel state information of domestic Wi-Fi network to monitor turnover activities.Unlike recent approaches placing the receiving antennas close to each other,this system introduces scattered placement,so that the spatial diversity of the receiving antennas can be fully exploited.As many as 8 typical sleep activities and 4 typical sleep postures can be recognized by this system.In order to obtain good recognize accuracy,a new error correction method based on the interrelation between consecutive activities is proposed to effectively correct the recognition error of CNN.Experiment results show that the system can significantly improve the recognition accuracy of both sleep activities and postures.In some cases,the recognition accuracy can reach as high as100%.