Research On Indoor Personnel Identity And Daily Action Recognition Technology Based On CSI

With the wide deployment of Wi-Fi devices in indoor environments,the recognition and classification technology of wearless devices based on Channel State Information(CSI)has gradually attracted people’s attention.Compared with traditional identification systems,the technology based on CSI has many advantages,which can avoid the high deployment cost,privacy invasion and application environment restrictions of traditional systems.In short,CSI-based device-free recognition technology can be widely deployed in a variety of environments and can work within Non-Line of Sight(NLOS),even through walls for recognition.These characteristics make the technology more flexible and applicable in practical applications.Identity identification plays an important role in daily life,such as a wide range of application scenarios,such as time check-in,account login and security verification.Aiming at the disadvantages of traditional human identification system,such as expensive equipment and invasion of privacy,this paper proposes a Wi-Fi based CSI human identification system.Firstly,the data information reflecting human gait characteristics is extracted from the improved wireless signal receiving equipment,and then the data is preprocessed and the key feature data is extracted.The improved Support Vector Machine(SVM)algorithm is used to realize the classification.In this paper,an iterative optimization of SVM parameter combination method based on Genetic Algorithm(GA)for Wi-Fi personnel identity recognition is proposed,and an adaptive adjustment training classifier is carried out to achieve the optimal identification rate.With the development and demand in human-computer interaction,medical monitoring,virtual reality,and security,CSI-based action recognition has become an important research topic and is used in many human-centered services and applications.Most of the existing action recognition studies only use the amplitude information in CSI data and ignore the phase information,and few system models use both amplitude and phase information for recognition.In addition,the field of machine learning and deep learning has made rapid development in recent years,especially Convolutional Neural Network(CNN)can better handle multi-channel data and improve recognition performance,Bi-Directional Long and Short Term Memory(Bi LSTM)can consider both forward and BiDirectional LSTM(Bi LSTM)can consider both forward and backward information,better capture the sequence semantics,reduce the dependence on the input sequence,and have better generalization ability.To address the above research status,this paper proposes a framework for an action recognition system based on CNN and Bi LSTM networks with combined amplitude and phase information,which uses a combined CNN and Bi LSTM network as classifier to investigate the system performance using both amplitude and phase values.In addition,the robustness of the system in different environments is verified with different data sets in different scenarios(conference room and lecture hall).