Research On Hand Gesture Recognition Based On 77GHz Millimeter Wave Radar

Human-computer interaction occupies an important position in the science and technology.Among them,hand gesture interaction has a very broad application prospect in the fields of smart home,somatosensory games and wearable devices with its natural and intuitive characteristics.Currently,more mature technologies of hand gesture recognition are based on visual and inertial sensors.However,the vision-based hand gesture recognition scheme is greatly affected by the lighting conditions.The inertial sensor-based hand recognition scheme needs to wear an additional sensor to the user's hand,which has great non-portability and seriously affects the user's experience.The radar-based hand gesture recognition system does not require the human body to wear additional sensors and is not affected by lighting conditions,can work normally around the clock,and the radar signal has good penetration.Compared with hand gesture recognition systems based on low-band wireless signals,millimeter-wave radar sensors are easy to miniaturize and can be embedded inside the device,greatly improving the integration and reliability of the device.At the same time,the millimeter wave signal is more capable of capturing small movements due to its extremely short wavelength,and can effectively recognize the fine movement of the finger.Based on the above background,this paper proposes a hand gesture recognition system based on 77 GHz millimeter-wave radar.The key technologies are deeply explored and a complete hand gesture recognition system is constructed.The main work of this paper is as follows:(1)Starting from the actual scene,the design of the radar signal is completed,the acquisition and preprocessing of the dynamic hand gesture data are also completed.Firstly,the signal of FMCW radar is designed according to the experimental scene and the predefined hand gesture action.Then,the effective hand motion information is processed from the echo signal to eliminate the influence of environmental noise.(2)Analysis and research on feature extraction of hand gestures.Dynamic hand gestures are the combination of hand and fingers movement in continuous time.Different dynamic hand gestures have different modulations on radar signals.Based on the classical time-frequency analysis method,this paper proposes the method of using short-time Fourier transform.MicroDoppler signatures characterizing the dynamic hand gestures are extracted from the echo signal to construct the set of hand gesture signatures.(3)Research on the classification model of hand gestures,and propose a shallow convolutional neural network classification model to identify the set of hand gesture signatures.The use of a non-deep structure in the classification network not only reduces the size of the model but also effectively improves the recognition speed of the sample,providing a possibility for future layouts in wearable devices and mobile devices.(4)The paper proposes the design of various experimental scenarios,explores the influence of different environmental factors on the accuracy of hand gesture recognition,and further proposes a convolutional neural network model based on the open parallel structure of Inception to identify the full scene gesture set,while maintaining the structure is simple and the recognition accuracy is effectively improved.According to the actual test scenario,the hand gesture recognition system based on 77 GHz millimeter-wave radar can effectively identify nine kinds of predefined common dynamic hand gestures under a small number of samples,which has good recognition accuracy and recognition speed.It provides a new idea for the subsequent research of human-computer interaction.