Research On Indoor Motion Recognition Based On WiFi Signal

Human motion recognition has broad application prospects,such as health care,smart home and so on.Recently,since different human body actions have different effects on WiFi signals,and Channel State Information(CSI)can record this change more accurately,CSI can be used for human motion recognition.Human behavior recognition based on WiFi non-contact method has the characteristics of not revealing user privacy and being unaffected by light intensity,so it will be an important supplement of video-based human behavior recognition technology and has broad application prospects.To this end,this paper designs and implements an indoor motion recognition system based on WiFi signal.The main work is as follows:(1)This paper designs and implements an indoor motion recognition system based on WiFi signal.This system includes four modules: data acquisition,data preprocessing,motion interval interception and data compression,and motion recognition.We successfully implemented our human body recognition system and performed data acquisition and performance testing on the system.(2)Propose an action interval interception algorithm based on double threshold.The original collected data contains non-action parts.In order to intercept the action interval more accurately,this paper proposes a double-threshold action interval interception algorithm based on mean absolute deviation(MAD).This algorithm first compares the MAD values of different windows by threshold to determine whether For meaningful waveforms,then further threshold comparisons are performed using small windows to more accurately extract data segments containing human motion.(3)Design two different motion recognition methods for final motion recognition,which are based on template matching method based on dynamic time warping and K-nearest neighbor and method based on convolutional neural network.The experimental results show that the average accuracy of 88% is achieved by using template matching recognition.The convolutional neural network achieves an average recognition rate of 95%.The experimental results show that the two methods are effective,and the feasibility of using the WiFi signal for human body recognition is illustrated.