Research On Ultrasonic Mid-air Gesture Recognition Method Based On Waveform Analysis

Mid-air gesture interaction is a kind of gesture interaction mode that can be operated empty-handed without touching the equipment.Because of its wide range of interaction,high flexibility and natural interaction experience,it is widely used in the fields of safe driving,medical treatment,robotics,experience games and so on.It has a good research prospect.Ultrasonic gesture data perception mode is not affected by light,skin color,electromagnetic interference and other factors applied in mid-air gesture interaction;The existing gesture interval detection mainly finds the start and end interval of gesture by setting threshold value,which is easy to be affected by the device error and environmental noise.At the same time,the eigenvalues extracted manually in mid-air gesture recognition are often complicated to calculate and difficult to represent different gestures,and the neural network algorithm that can extract eigenvalues independently is difficult to obtain the long-term dependence of mid-air gesture.Therefore,in this paper,a gesture data perception method based on ultrasound is adopted to extract ultrasonic delay by locating the arrival time of palm echo,further extract gesture track information and realize data normalization and dimensionality reduction,extract effective feature vectors,and construct the hidden Markov model and the mid-air gesture recognition model of short and long time memory network respectively.The main research contents and contributions of this paper are summarized as follows:(1)The overall architecture of ultrasonic mid-air gesture recognition system is designed,and the hardware platform structure and data acquisition process of the system are introduced.Based on the principles of interaction process,user experience and system performance,five gesture categories were designed and the original data set was constructed,which provided data support for ultrasonic mid-air gesture recognition research.(2)The direct noise signal and environment noise signal were removed by noise reduction and smoothing algorithm,and the palm echo signal was highlighted.The ultrasonic time delay extraction algorithm is proposed to locate the arrival time of palm echo,and the three-dimensional spatial information of echo is used to extract the gesture trajectory information,which provides support for the subsequent feature vector extraction.According to the mid-air gesture designed in this paper,the gesture frame segmentation algorithm is proposed to achieve the data normalization and dimension.(3)This paper improved the recognition effect of mid-air gesture from the two perspectives of feature vector extraction and model construction.The feature vector was extracted by echo flight distance difference information,and its correlation with gesture labels was quantified by maximum information coefficient.The effectiveness of the feature vector proposed in this paper was verified by experiments.From the perspective of model,a hidden Markov mid-air gesture recognition model based on time sequence state transition was constructed by using the extracted feature vector.The experimental results show that the recognition rate reaches 83.75%.In order to extract the long-term dependence of gestures,a mid-air gesture recognition model based on short and long time memory network was constructed and optimized by Adam algorithm.The experiment showed that the recognition rate reached 89.8%,which was6.05% higher than that of the hidden Markov model.