The Research Of Indoor Human Activity Perception Based On Wi-Fi

In recent years,with the development of all kinds of applications of Internet of Things,the wireless activity sensing technology for indoor human body is responding to the advocacy of smart life,and it is inextricably linked with people's daily life.Such as indoor localization navigation,security monitoring and medical monitoring.It has brought great convenience to people's indoor life,and also provided potential opportunities for the development of various location services,helping smart life to move from concept to reality.Wi-Fi technology has attracted much attention in the field of wireless activity sensing because of its high transmission rate and low deployment cost.Therefore,it has application value for Wi-Fi-based indoor human activity sensing scheme research.According to the two main stages of indoor human body activity sensing:location sensing and behavior sensing,this paper proposes an indoor localization scheme based on RSSI propagation model and a human behavior recognition scheme based on CSI,which uses different granularity of wireless signals to meet the demand of indoor human activity sensing.For the demand of location sensing,the parameter estimation and correction method of the RSSI ranging model is introduced firstly.Secondly,from the perspective of non?geometric positioning and bionics-based theory,an improved localization algorithm,DW-SAPSO localization algorithm,is proposed.Verify it in different actual scenarios.The experimental results show that,due to the introduction of simulated annealing and decreasing-inertia weight mechanism,the improved algorithm has improved the positioning accuracy and convergence performance in open and semi-closed experimental scenes and fully closed experimental scenesFor the demand of behavior sensing,it is first clarified that the RSSI's characterization of the multipath channel is not as good as the CSI of the resolution at the subcarrier scale.Therefore,CSI is selected to achieve more fine-grained activity perception.Secondly,the amplitude and phase preprocessing of the CSI raw data is performed.Two mechanisms for reducing computational overhead are proposed based on the moving variance threshold and wavelet transform.The Doppler power spectrum is proposed in feature extraction.Finally,the combined classifier weighted voting method is proposed for behavior recognition.The proposed scheme is verified in different practical scenarios.The experimental results show that the overall behavior recognition accuracy of the proposed method is higher than that of a single classifier in semi-closed corridor and fully closed laboratory scenarios.In addition,random forest is most sensitive to the characteristics of falling and has the highest recognition accuracy of falling.