Design And Implementation Of A Low Computing Power Device Based Wifi Wireless Sensing System

With the development of Internet of Things technology,smart home devices have been everywhere,and using gestures to interact with smart home devices is an emerging interaction method.At present,the technology of using wireless signals to recognize gesture actions includes millimeter wave,WiFi,etc.Among them,WiFi is easier to deploy and lower in price,so it has attracted much attention.Existing related research mainly focused on increasing the accuracy of gesture recognition,and less attention is paid to the practical application in smart home scenarios.This article mainly focuses on using device with low computing power in the smart home scenario to realize gesture recognition with WiFi signals,and further interact with smart home devices.This paper includes:(1)A gesture dataset collecting platform based on the device with low computing power was built,and thousands of samples of interactive gestures were collected.(2)Designed a gesture detection algorithm based on dynamic threshold,which can continuously detect the input CSI data stream with low delay,and extract the data segment containing gestures;designed an efficient signal preprocessing method,which can quickly remove the outliers in the signal and make it smooth.This method reduces the time consumed by preprocessing and improves the real-time performance of gesture recognition.(3)A lightweight gesture recognition model is proposed,and the depth-separable convolution is used to reduce the amount of calculation required for the convolution operation of the traditional convolutional neural network.The extracted gesture action data is processed into a image,and the gesture action recognition is realized by using the convolutional neural network combined with the attention mechanism.(4)Built an interactive system to realize the display of gesture recognition results,interact with photo browser and video player,and provide interfaces for further access from other smart home devices.Finally,the function test and performance test are carried out and the performance is compared with other recognition models.The accuracy of our system can reach 98 percent on four kinds of gestures.