Research On Gesture Recognition Technology Based On Micro-Doppler Radar

Gesture recognition based on micro-Doppler radar is realized by classifying microDoppler features extracted from electromagnetic waves which can captures the gestures of human targets.It is a steady and non-contact recognition method,which has the advantages of being less restricted by environmental conditions,less risk of privacy leakage,and easy to deploy to the system.Therefore,it can be widely used in audio-visual entertainment,smart home,and assisted driving.In this paper two gesture recognition algorithms are developed from micro-Doppler features of gesture actions.(1)The technical principles concerning to gesture recognition are reviewed,followed by an introduction of The principle of micro-Doppler radar,the analysis and extraction of microDoppler features,and the structure and recognition principle of classification models are studied.All of the theories listed above are the preliminaries of gesture recognition.(2)The collection and preprocessing of gesture signals are studied,including the design of gestures types to be collected and experimental conditions.A gesture database with a sample size of 8000 is established through micro-Doppler radar.The Short-Time Fourier Transform(STFT)of radar echoes from hand gestures is applied to product the time-frequency spectrograms of hand gestures.The background clutter in the time-frequency spectrogram is removed using the difference in signal distribution intensity,which provides support for the subsequent feature extraction and classification.(3)A new recognition method based on Histograms of Oriented Gradients(HOG)features is developed for the characteristics of time-spectrogram considering artificially extracting of features.The principle of edge detection is used to extract the micro-Doppler features of gesture actions.Then Support Vector Machine(SVM)is used as the classifier for classification and recognition of feature data which is set as input.(4)A lightweight recognition network is designed using deep separable convolution and inverse residual blocks based on Convolutional Neural Network(CNN)to adapt automatically extracting features conditions.It can not only ensure recognition accuracy but also reduces the number of parameters in the network and the performance requirements of the hardware.Experimental results based on measured data show that the two gesture recognition algorithms proposed in this paper can effectively extract the micro-Doppler features of gesture actions and have high recognition accuracy.