摘要
为了研究调频连续波(FMCW)毫米波雷达在手势识别算法中的应用前景,本文通过使用77 GHz毫米波雷达获得了手势动作的回波,在手势动作回波的多帧距离多普勒图中提取了手势动作的距离−时间谱和多普勒−时间谱,并在此基础上建立了二维数据集。使用全局平局池化代替了扁平层与全连接层,构建了双通道的卷积神经网络对手势动作进行了分类,实现了对推、拉、推拉、挥手等7种手势的分类识别,测试集准确率可达99%,相较于单通道卷积神经网络提升了2%~4%。
In order to study the application prospect of frequency modulated continuous wave(FMCW)millimeter-wave radar in gesture recognition algorithm,this paper uses 77GHz millimeter-wave radar to obtain the echo of gesture action,and extracts the distance-time spectrum and Doppler-time spectrum of the gesture action from the multi-frame range Doppler map of the gesture action echo.On this basis,a two-dimensional data set is established.By using global average pooling instead of the flat layer or fully connected layer,a two-channel convolutional neural network is constructed to classify gesture actions.The recognition of 7 kinds of gestures,such as push,pull,push and pull,and wave,is realized,and the accuracy of the test set is up to 99%.The accuracy of dual-channel convolutional neural network is 2%~4%higher than that of single channel convolutional neutral network.
作者
陈涛
张法桐
刘子铭
CHEN Tao;ZHANG Fatong;LIU Ziming(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)
出处
《应用科技》
CAS
2021年第6期23-27,共5页
Applied Science and Technology
