Research On Gesture Recognition Algorithm Based On Millimeter Wave Radar

Gesture is an important way to realize human-computer interaction,users can easily control the electronic equipment according to different gestures.In recent years,with the relevant research on gesture recognition,millimeter wave radar equipment has gradually attracted people’s attention.Millimeter wave radar has a high resolution due to its short wavelength and bandwidth,and it is not affected by illumination conditions,good privacy protection,low power consumption and high portability.Compared with traditional gesture recognition methods,gesture recognition system based on millimeter wave radar has a wider application prospect.However,there are still some problems in the current research on gesture recognition based on millimeter wave radar,such as poor recognition effect on micro gestures,low accuracy of gesture recognition and complex input data.In this thesis,based on frequency-modulated continuous wave millimeter wave radar,the following research is carried out:(1)In order to solve the problem of complex data processing when multi-feature input and poor recognition effect when single feature input,this thesis proposes a neural network based on Res NetLong Short Term Memory,In terms of signal processing,the gesture recognition algorithm of Res NetLSTM(attention mechanism)obtains range-Doppler image RDM by extracting the distance and velocity characteristics of the original radar signal,and uses it as the input of the deep learning network,avoiding the complex network input algorithm which is relatively complex and easy to cause data redundancy and other problems.In terms of network architecture,In this thesis,the relevant features of the residual network are combined to prevent some useful information hidden in the data from being ignored,and the residual attention mechanism is introduced into the network to improve the network’s attention to the gesture features and avoid the impact of invalid features on the recognition accuracy.Furthermore,the timing features are processed in combination with the long and short term memory network to ensure the recognition accuracy of the network with single input features.Experimental results show that the proposed algorithm has certain advantages in recognition accuracy compared with other existing gesture recognition algorithms of millimeter wave radar.(2)Aiming at the problems of low recognition accuracy caused by single feature input of single feature gesture recognition algorithm and poor recognition effect on micro gestures,this thesis proposes a gesture recognition algorithm based on multi-feature fusion to preprocess the collected gesture echo signal to suppress clutter and remove static interference.The speed information,distance information and Angle information in the process of gesture movement are extracted and processed by combining the MUSIC algorithm.Range-doppler image RDM and range-angle image RAM of gesture trajectory are drawn.The multi-feature fusion network built by 3DCNN and CNN is used to process the two kinds of images respectively and input them into the network.The deep learning network was used to learn the hand feature information in the process of gesture movement,and then the gesture classification was completed according to different feature information.Through the relevant experiments on the self-built data set and the public data set,it is proved that the gesture recognition algorithm proposed in this thesis has better recognition effect compared with other existing gesture recognition algorithms based on millimeter wave radar in terms of gesture recognition accuracy and micro-motion recognition.