Research On WiFi Signal Gesture Recognition Based On Transfer Mechanism

With the rapid development of wireless communication technology,people find that wireless signal not only has the function of communication,but also has the ability to sense human activities in the environment.With the popularity of WiFi devices,activity recognition(AR)algorithms based on WiFi come into the sight of a large number of researchers.Especially after the fine-grained channel state information(CSI)in WiFi signal can be successfully extracted,this research direction is more popular.Compared with the traditional image-based and sensor-based perception recognition technology,WiFi based human activity recognition technology has many advantages,which can make up for the disadvantages of the traditional perception recognition system,such as:the dependence of image-based recognition system on light,the danger of visual dead angle and user privacy leakage;the intrusion of sensor-based recognition system to users,the risk of transmission,the expensive sensor equipment.From the perspective of indoor human gesture recognition based on wireless signal,this thesis studies the following three aspects:1.This thesis first introduces human gesture recognition technology based on traditional machine learning,then designs and configures the data acquisition environment,processes and applies data to the actual system,and finally realizes the smart home gesture control system based on machine learning.2.In order to solve the problem that the perceptual recognition system can not be transfered,this thesis proposes an ID based transfer mechanism,which uses conditional instance normalization(CIN)function to learn and map the identity information of people in source domain and target domain,and uses convolutional neural network as the backbone classification network to realize the human gesture recognition system based on cross ID transfer mechanism.The experimental results show that the average recognition rate can reach 98%in source domain and 95%in target domain.3.In order to solve the difficulty of experimental data acquisition,inspired by the idea of image style transfer,this thesis takes Cyclegan and Stargan as the generation network architecture,studies the data enhancement technology of channel state information based on WiFi signal,and designs the human gesture activity data enhancement network system based on Cyclegan and Stargan.The experimental results show that the data augmentation system can realize the purpose of expanding data set.