Research On Gesture Recognition And Authentication With Wi-Fi Signals

With the continuous development and popularization of wireless communication and Internet of Things devices,Wi-Fi wireless signal access has been widely used in the home,work,and daily entertainment environment.In addition to the normal communication function,Wi-Fi signal can also be used for human gesture recognition and identification,identity authentication,and other aspects.Compared with the traditional human gesture recognition technology(such as video surveillance,etc.),the human gesture perception and recognition based on Wi-Fi signal has the advantages of low cost,non-line-of-sight,privacy protection,and so on.In this paper,there are two problems in the current indoor human gesture recognition and authentication based on Wi-Fi signals: Given the problem that the user gesture recognition is greatly affected by the orientation of users and the small number of participants in user training in the machine learning model,the research of channel state information based on Wi-Fi signals is carried out to realize the robustness of user gesture recognition with different orientations and multi-user joint gesture recognition under federated learning.Specifically,this paper mainly carries out the following two aspects of work:(1)Aiming at the problem that the current Wi-Fi gesture recognition technology requires users to perform gestures in a fixed orientation,a Wi-Fi gesture recognition system based on deep learning network classification is designed and proposed.Using the phase difference of channel state information of users' gestures with different orientations,using robust PCA(r PCA)to de-noising,reducing the influence of body and other parts on gestures,using Gram Angle field to carry out time sequence coding of the first principal component after robust PCA,and designing a multi-channel deep neural network classification and recognition model.To realize robust gesture recognition and authentication for different orientation recognition of user gestures.(2)For the current Wi-Fi signal gesture recognition technology combined with the machine learning method,a large number of users need to participate in the construction of the training set,to improve the robustness of the model for new users.In this paper,a federated learning method based on parameter matching is proposed to effectively solve the problem that a small number of users can participate in a single environment to achieve the strong robustness of the model to new users.The experimental results show that when seven users participate in model training in two different rooms,the system can achieve an average accuracy of 90.4% for new users' gestures.