Research On Gesture Recognition Based On Millimeter Wave Radar

The gesture recognition based on millimeter wave radar is to use electromagnetic waves to obtain the motion characteristics of the human body's dynamic gestures,and to realize the automatic recognition of gestures.As a non-contact identification method,it can achieve simple and efficient information interaction,and has a wide range of application needs in medical,automotive,entertainment,etc.In recent years,it has become a research hotspot at home and abroad.High accuracy recognition effect is the key to gesture recognition research.This article takes the different characteristics of gestures as input,combines supervised and unsupervised methods,and focuses on three gesture classification algorithms,and uses millimeter-wave radar to collect dynamic gestures for verification.The specific contents are as follows:1.Taking gesture single feature as input,a gesture recognition method based on tandem architecture network is proposed.In the existing gesture classification algorithm based on convolutional neural network(CNN),this thesis finds that the recognition rate of gestures with similar features is low,which can easily lead to uneven accuracy of various gestures.In response to this problem,based on the existing CNN gesture classification results,this thesis screens out similar gestures and designs different algorithms for reclassification according to their characteristics,which can effectively improve the recognition rate of similar gestures.The two-layer classification network constitutes the serial architecture network of this thesis in a cascading manner.The experimental results show that,compared with the existing CNN method,the algorithm has high average accuracy and the recognition rate of various gestures is relatively balanced.2.Taking multi-features of gestures as input,an unsupervised gesture recognition method based on PCA and K-means is proposed.Existing supervised radar gesture algorithms are all learning and classifying on a large number of labeled high-dimensional gesture sample sets,which has the problems of large amount of calculation and long time-consuming for labeled samples and training models.This thesis proposes to use PCA to reduce feature dimensions and data volume,and then use K-means method without labeling to perform clustering without repeated training and learning.At the same time,the K-means method of random initial clustering centers is easy to locally converge and difficult to apply to radar data.This thesis proposes to improve the initial center based on the density offset,which not only can globally converge,but also is found to be more suitable for radar after verification Gesture data.Experimental results show that this algorithm saves time,has good clustering effect and low computational complexity.3.Taking gesture multi-features as input,a supervised gesture recognition method based on multi-feature fusion is proposed.Most of the existing supervised algorithms use single features and limited representation of complex dynamic gestures.In response to this problem,this thesis takes various information such as distance,speed,and angle of dynamic gestures as input,and makes full use of gesture information.At the same time,the traditional CNN recognition network is not good at extracting the timing features of dynamic gestures.In this thesis,a multi-input CNN-LSTM supervised network is designed to fully extract and integrate the timing features of dynamic gestures.The experimental results show that,compared with the mainstream gesture recognition algorithm based on a single feature,this algorithm improves the problem of low recognition rate and obtains an average accuracy rate of up to 96.4%.The three methods proposed in this thesis have been verified by a large number of measured radar data.The results show that these methods start from different focuses,use different features and classification algorithms,and effectively improve the existing imbalanced recognition rate,large calculation and single feature in the existing radar gesture recognition process,and achieve a variety of high efficiency accurate radar gesture recognition algorithm.