| In recent years,As science and technology develops by leaps and bounds,the improvement of precision manufacturing technology and the reduction of manufacturing costs,radar has many application scenarios and values from most of the past applications in the military field to the present in the civil field.Compared with other traditional sensors,radar has unique advantages in the field of hand gesture recognition applications.Millimeter wave radar stands out in many systems of radar by virtue of its small size,strong penetration ability,high Doppler frequency shift and large bandwidth.The basis of gesture recognition based on millimeter wave radar is that the movement of gesture changes the echo signal received by the radar and extracts the key information in the gesture echo signal.This paper is based on the acquisition of hand gesture echo signals by millimeter wave radar,and carries out recognition and classification through a series of signal processing algorithms,target tracking algorithms,feature screening algorithms and machine learning algorithms.The main work of this paper is as follows:2.Aiming at the problem that the actual hand signal acquisition environment contains many kinds of target interference,such as static interference and dynamic interference,which affect the effect of subsequent gesture recognition,this paper proposes a solution based on the target tracking algorithm such as Unscented Kalman filter.First of all,build the AWR1843 millimeter wave radar data acquisition platform,preprocess the collected hand signal,improve the signal-to-noise ratio and eliminate the background noise;Secondly,three kinds of gesture motion models are established by using the unscented Kalman filter algorithm and the motion trajectories of various gestures are predicted.After that,the predicted gesture data information and the actual gesture motion information are classified by using the K-nearest neighbor algorithm,and the parameter weights that do not belong to the gesture trajectories are reset to zero,so as to eliminate the interference of gesture targets;Finally,input the range-doppler characteristic spectrum of the target like interference that has been eliminated into the convolution neural network for recognition,and compare and analyze the experimental results of inputting the characteristic spectrum of the target like interference that has not been suppressed into the convolution neural network,and compare with the experimental results of the target like interference suppression algorithm proposed in other documents.On the one hand,the target-like interference suppression algorithm proposed in this paper effectively improves the accuracy of gesture recognition and accelerates the convergence speed.On the other hand,this paper has a very good recognition effect for confusing gestures such as clockwise rotation and counterclockwise rotation,with an average recognition rate of 98.03%.It shows that the algorithm proposed in this paper has good performance in hand gesture recognition in the scene with more interference of similar objects.3.Aiming at the problems of inaccurate angle estimation results and poor classification results caused by redundant gesture features in millimeter wave radar gesture angle extraction,this paper proposes a solution based on source estimation algorithm and feature recursive filtering algorithm based on support vector machine.First of all,on the basis of the millimeter wave radar hand signal acquisition platform built in chapter 3,the hand signal is preprocessed;Secondly,the minimum description length criterion in the source estimation algorithm is used to optimize the multi-signal classification angle estimation algorithm to improve the accuracy of angle estimation;Furthermore,six kinds of gesture motion features are fully mined,and 17 kinds of features are proposed to describe the gesture motion trajectory in detail,and the redundant features are filtered by the feature recursion algorithm based on support vector machine to select more useful gesture features;Finally,support vector machine is used to classify six kinds of gestures,and the classification results of gesture features without angle information are compared with the classification results of gesture features with angle information.At the same time,the results are compared with the experimental results of gesture feature filtering algorithm proposed by other literatures.On the one hand,the accuracy of hand gesture classification results with angle information is higher than that without angle information,which verifies the necessity of the angle optimization algorithm proposed in this paper;On the other hand,the feature filtering algorithm proposed in this paper has a better recognition effect for gesture recognition,with an average accuracy of 98.77%,which shows that the algorithm proposed in this paper has a good performance for inaccurate angle estimation and feature redundancy. |
