Research On Key Technologies Of Gesture Recognition With Wi-Fi Using Behavior Trajectory

With the development of computer and communication technology,the application field of human-computer interaction technology in daily life is expanding constantly.Traditional sensing methods cannot achieve high precision activity sensing on the premise of device portability and privacy.At present,perception recognition technology based on commercial Wi-Fi can provide users with contactless perception services.However,due to the impact of hardware noise of commercial Wi-Fi equipment,there are still great challenges in positioning error,tracking accuracy and recognition ability.Aiming at the above challenges,this thesis studies the channel state information(CSI,Channel Status Information),the key technology of gesture recognition based on behavior trajectory.Firstly,as the basis of gesture trajectory recognition,a human hand moving distance measurement system based on CSI ratio features is designed.In this system,CSI ratio features between adjacent antennas are used to eliminate phase noise such as packet detection delay and carrier frequency offset in the Wi-Fi system,reduce amplitude noise generated by power amplifier in the Wi-Fi system,extract motion-related information in CSI,and calculate the moving trajectory of the target.The simulation results show that the average moving distance measurement error of the system is 4.9cm.Secondly,in order to improve the accuracy of gesture trajectory tracking,a gesture trajectory tracking scheme based on Doppler velocity smoothing algorithm is proposed to solve the problem of trajectory incoherence estimated by the above system.In this scheme,the path change rate is extracted from Doppler spectral features triggered by target activity.On this basis,the path change rate is modeled as a real-time velocity and acceleration function from the point of view of physics to smooth the velocity curve.Finally estimate the trajectory of the target.Simulation results show that the proposed method can effectively estimate the activity trajectories of targets from CSI data.In the measured data,the average error of gesture trajectory tracking is 2.3cm.Finally,a gesture recognition method based on dual flow integrated activity recognition neural network model was proposed to solve the problem of inaccurate gesture recognition caused by trajectory tracking errors.Different from the traditional pure trajectory recognition scheme,this method fuses Doppler spectrum and motion trajectory at the feature level,which can compensate the moving details of the target.On this basis,considering the generalization ability of neural network model,data enhancement and loss function strategy are combined to improve the accuracy of cross-scene identification of the model.The simulation results show that compared with the pure trajectory recognition scheme,the overall recognition accuracy of the proposed method is improved by 4%.