Low-Cost And High-Precision Cross-Target Gesture Recognition System With Wi-Fi And Video

Wireless sensing technology based on Wi-Fi signals has the advantages of low-cost,easy deployment and good privacy protection.Among them,gesture recognition has gradually become a research hotspot in the field of Internet of Things due to its wide demand in cutting-edge applications such as smart home and automatic driving.However,the existing gesture recognition technology is faced with the problem of low accuracy or even failure in cross-target recognition,that is,due to the difference in height,body shape and gesture habits of different users,the performance of gesture recognition for new users using the trained recognition model is greatly reduced.To solve this problem,this paper proposes a low-cost cross-target gesture recognition system Wi-Ges with video and Wi-Fi.Its main contributions and research contents are as follows:(1)Video based multi-user gesture Wi-Fi data generation methodTo solve the failure problem of cross-target recognition model,the traditional method generates a large amount of user data through deep learning network,but it has poor robustness and low interpretability.To this end,the paper put forward the use of video data generated multi-user gestures Wi-Fi data,by means of a small amount of training to user gesture video data extraction and high diversity of characteristics,build a network model to generate a large number of different users of gestures Wi-Fi data,not only reduces the cost of training data collection,and improve the robustness and generalization ability of the system.(2)Cross-target gesture recognition model based on Unet architectureGesture recognition is carried out based on the above generated multi-user Wi-Fi data,and a cross-target gesture recognition model based on the Unet architecture is proposed in this paper considering the influence of multi-path in the real environment.Specifically,the gesture feature mapping relationship between the generated Wi-Fi data and the real Wi-Fi data in the multipath environment is established by introducing the Unet framework,and the multipath influence is filtered out to achieve high-precision cross-target gesture recognition.Through a large number of experiments,this paper identifies 20 different users and 10 gestures.Finally,the cross-target recognition accuracy of Wi-Ges on 5 and 10 gestures reached 89.30% and 84.80%,respectively.