Research And Implementation On Gesture Recognition Method Based On High Frequency Acoustic Of Smartphone

The large number of applications of the new APP has put forward a wider demand for interactive forms.At present,the traditional forms of interaction based on cameras and sensors are inadequate.The former is easy to expose user privacy,while the latter user experience needs to be improved.Therefore,more and more researchers are turning their attention to how to use the acoustic wave of smart devices to perform action interaction.The current acoustic gesture recognition methods are mainly based on four types of methods: recording sound,doppler shift,phase difference,and correlation between the transmitted and received signals.The low frequency recording signal is easily affected by the noise in the environment,and the non-friction gesture cannot be recognized.The method based on doppler shift can only recognize simple gestures.The method based on phase difference can only obtain the change of relative position,which requires the user's gesture to be completed in one stroke.The method based on sending and receiving signal correlation can only obtain the change of absolute position,which requires the rest of the body to be immobile,otherwise the moving path of the target object cannot be distinguished.In order to solve the limitations of the above methods,this paper proposes an recognition method based on high-frequency acoustic waves named Acoustic Gesture.High-frequency sound waves are used to avoid the interference of low-frequency environmental noise,and deep-learning training models are used to recognize complex gestures,while eliminating the limitation that the above methods require the rest of the body to be immobile.The main research contents are as follows:(1)In order to solve the problem of pre-classification of dynamic and static gestures and avoid the large delay caused by the frequency domain feature pre-classification,a signal extraction method based on the time domain periodic signal peak and variance was proposed,and a dynamic threshold selection method was proposed to realize the online update threshold.(2)In order to solve the complex gesture recognition and cancel the limitation of the rest of the body's immobility,a method based on high-frequency sonic gesture recognition was studied.The different effects of dynamic gestures and static gestures on the channel were analyzed.Time-frequency features were used to train dynamic gesture models,and spectrum features were used to train static gesture models.(3)Design and implement a gesture recognition system based on high-frequency sound waves of Android smartphones.A full experimental evaluation of Acoustic Gesture was conducted,comparing different users,different scenarios and related research.The experimental results show that Acoustic Gesture has low hardware requirements,is not affected by external noise,and has small restrictions on user gestures,and has high robustness,high recognition accuracy,using Convolutional Neural Network feature extraction,the accuracy rate is increased by about 5%,and the accuracy rate of identifying letters,numbers,dynamic gestures,and static gestures reaches 97.4%,95.5%,99.9%,97.4%.