基于微多普勒特征和深度学习的人体动作识别  

Human Motion Recognition Based on Micro-Doppler Features and Deep Learning

作　　者：钟滢洁 李秋生[1,2] ZHONG Yingjie;LI Qiusheng(Research Center of Intelligent Control Engineering Technology,Gannan Normal University,Ganzhou 341000,China;School of Physics and Electronic Information,Gannan Normal University,Ganzhou 341000,China)

机构地区：[1]赣南师范大学智能控制工程技术研究中心,江西赣州341000 [2]赣南师范大学物理与电子信息学院,江西赣州341000

出　　处：《赣南师范大学学报》2022年第6期37-42,共6页Journal of Gannan Normal University

基　　金：国家自然科学基金资助项目(61561004);江西省教育厅科学技术研究项目(GJJ201408)。

摘　　要：针对基于光学和红外视频数据的人体动作识别受环境影响大及传统机器学习分类方法特征提取复杂的问题,提出基于雷达微多普勒频谱图和深度学习模型的人体动作分类识别方法.首先,搭建77GHz毫米波雷达数据采集系统.其次,开展人体行为回波数据预处理,有效提取微多普勒信息,得到人体行为二维距离多普勒图像数据集.最后,以距离多普勒谱图作为网络的输入样本,设计3层卷积与池化操作构建特征空间完成4种不同人体动作识别的仿真实验.实验结果表明,与现有的人体动作识别方法相比,将77GHz调频连续波雷达回波进行距离多普勒处理与CNN结合能够实现对日常人体动作的有效识别,识别准确率可达98.91%,优于传统方法.Aiming at the problems that human motion recognition based on optical and infrared video data is greatly affected by the environment and the feature extraction of traditional machine learning classification methods is complex,a human motion classification and recognition method based on radar micro-Doppler spectrogram and deep learning model is proposed.First,the data acquisition sys-tem of 77GHz millimeter wave radar is built.Secondly,the preprocessing of human behavior echo data is carried out to extract micro-Doppler information effectively,and the two-dimensional range Doppler image data set of human behavior is obtained.Finally,with the distance Doppler spectrum as the input sample of the network,three-layer convolution and pooling operation are constructed to con-struct the feature space to complete the simulation experiment of four different human motion recognition.The experimental results show that,compared with the existing human movement recognition methods,the combination of 77GHz frequency modulated cw ra-dar echo with range Doppler processing and CNN can realize the effective recognition of daily human movements,and the recognition accuracy can reach 98.91%,which is better than the traditional method.

关 键 词：毫米波雷达 人体动作识别 微多普勒 卷积神经网络 

分 类 号：TN957.51[电子电信—信号与信息处理]