Gesture Recognition Based On Multi-feature Fusion Of Millimeter Wave Radar

Millimeter-wave radar technology has been widely used in the rapid development of wireless sensor technology.Millimeter-wave radar is used in smart factories,autonomous driving and smart buildings.As a wireless sensing device,radar can not be affected by light and does not reveal privacy compared with other devices,So it’s very widely used in human behavior detection.Gesture is widely used in people’s production and life as well as human-computer interaction.Light or occlusions are very easy to interfere with gesture recognition,so it can be applied in very narrow scenes.But radar signals are not affected by jamming and operate over a wide range of spectrum,allowing them to recognize subtle movements such as dynamic gestures.So the use of millimeter-wave radar gesture recognition has important practical significance and research value.At present,the use of millimeter-wave radar to sense dynamic gestures is a hot research topic.Millimeter-wave radar has the advantages of high range measurement,velocity measurement and angle measurement,which is very suitable for the detection of human gestures.For different dynamic gestures,it is very important to select different gesture features in the data preprocessing stage,which can significantly improve the recognition effect of dynamic gestures.Based on this,this thesis selects different gesture features for three kinds of dynamic gestures in different experimental scenes,and uses different classification models to complete the classification and recognition of gestures.The main work contents are as follows.(1)This thesis proposes a gesture recognition method based on range-doppler features,which can recognize coarse-grained dynamic gestures.This method analyzes range-doppler information in detail,and the range-doppler features that eliminate filtering interference can fully describe coarse-grained dynamic gestures.While improving the utilization rate of radar echo signal,it also has better gesture recognition effect.(2)A multi-scene digital gesture recognition method based on three-dimensional gesture feature set is proposed,which can recognize fine-grained digital gestures.In experiments in different scene domains(environment,location,direction and people),it is difficult to describe the gesture movement information in detail.So use the distancetime diagram,the doppler-time diagram,and the angle-time diagram to describe the digital gesture.And by means of Convolutional Neural Network CNN,the threedimensional gesture features can be trained and recognized,which can achieve a good recognition effect.When three-dimensional gesture features are used to recognize dynamic gestures,the recognition range of dynamic gestures is improved,and then the feasibility of millimeter wave radar to recognize fine-grained gestures is illustrated.(3)In this thesis,a method based on micro-doppler features is proposed to recognize the local gestures in the process of writing characters in the air.In order to solve the problem that it is difficult to extract gesture features when writing characters in the air,micro-doppler features are used to describe local gesture movements,and written letter gestures are recognized by the algorithm model integrated by CNN and Long Short-Term Memory.This method can recognize local gesture motion in the scene of writing characters in the air,which shows the effectiveness of micro-doppler features in recognizing local motion.It not only improves the accuracy of gesture recognition,but also enriches the method of gesture recognition.It further explains the importance of selecting gesture features when using millimeter wave radar to recognize gestures.