Research On Micro Gesture Recognition Method Of Millimeter Wave Radar

As a human-computer interaction method,gesture recognition technology based on millimeter-wave radar has the advantages of all-weather,no privacy exposure,small size and easy deployment.It has a wide range of application prospects in smart medical care,smart home and smart cockpit..Background clutter suppression,micro-gesture feature extraction and micro-gesture classification are the key links in micro-gesture recognition.Focusing on the problem of micro-gesture recognition of finger joints,this thesis mainly studies the multidimensional feature gesture recognition methods of existing categories and the incremental learning method of mnemonics for new types of data.The main contents are as follows:1.To suppress the background clutter of the millimeter-wave radar micro-gesture,on the basis of the range-dimension suppression,the Doppler-dimension fourth-order feedback filtering method is used to effectively suppress the background clutter of the micro-gesture.2.Aiming at the problem of effective extraction of micro-gesture echo features,the parameters of gesture echo are analyzed,and it is proposed to extract features from the distance,Doppler and angle of dynamic gestures,and obtain the multi-dimensional feature map of gestures.3.Aiming at the recognition problem of new micro-gesture categories,a microgesture recognition method based on mnemonic incremental learning is studied.The problem of catastrophic forgetting of existing categories in incremental learning is alleviated by sample parameterized training,and the new gesture category is realized.accurate identification.4.Aiming at the recognition problem of new micro-gesture categories,a microgesture recognition method based on mnemonic incremental learning is proposed.The use of sample parameterized training alleviates the problem of catastrophic forgetting of existing categories in incremental learning,and realizes new Accurate recognition of gesture categories.The above research methods have been experimentally verified through the measured micro-gesture dataset.The results show that the method proposed in this thesis can recognize micro-gestures efficiently and accurately,and can also maintain high recognition accuracy for new gesture categories.