查询结果:   蒋留兵,魏光萌,车俐.基于卷积神经网络的雷达人体动作识别方法[J].计算机应用与软件,2019,36(11):168 - 174,234.
中文标题
基于卷积神经网络的雷达人体动作识别方法
发表栏目
人工智能与识别
摘要点击数
354
英文标题
HUMAN MOTION RECOGNITION METHOD BY RADAR BASED ON CNN
作 者
蒋留兵 魏光萌 车俐 Jiang Liubing Wei Guangmeng Che Li
作者单位
桂林电子科技大学广西无线宽带通信与信号处理重点实验室 广西 桂林 541004 桂林电子科技大学计算机与信息安全学院 广西 桂林 541004    
英文单位
Guangxi Wireless Broadband Communication and Signal Processing Key Laboratory, Guilin University of Electronic Technology, Guilin 541004, Guangxi, China School of Computer Science and Information Security, Guilin University of Electronic Technology, Guilin 541004, Guangxi, China    
关键词
人体动作识别 超宽带雷达 深度学习 卷积神经网络
Keywords
Human motion recognition UWB radar Deep learning CNN
基金项目
国家自然科学基金项目(61561010);广西自然科学基金项目(2017GXNSFAA198089);广西重点研发计划项目(桂科AB18126003,AB16380316)
作者资料
蒋留兵,研究员,主研领域:雷达信号处理。魏光萌,硕士生。车俐,高级实验师。 。
文章摘要
利用雷达来识别人体动作对环境要求较低,且避免了摄像头带来的的隐私问题。针对这种需求,提出一种基于超宽带雷达和深度学习算法的人体动作识别方法。利用超宽带雷达的高距离分辨力,并针对人体动作的动态特性,提取出人体目标的距离-时间二维特征,弥补单一距离特征的不足。针对特征图采用一种经过优化的卷积神经网络进行识别。通过SIR-20高速探地雷达平台进行数据采集,对8种不同的人体动作进行识别,最终达到了平均99.2%的正确识别率,验证了该方法的可行性和有效性。
Abstract
The use of radar to recognize human movements has low environmental requirements and avoids privacy issues caused by cameras. Addressing this need, this paper proposes a human motion recognition method based on Ultra-Wide-Band(UWB) radar and deep learning algorithm. Considering the high distance resolution of UWB radar and the dynamic characteristics of the human body movement, the method extracted the distance-time two-dimensional feature of the human body target. It covered the shortage of the single distance feature. For feature maps, an optimized convolution neural network was adopted to perform recognition. Data collection was carried out on the SIR-20 high-speed ground-penetrating radar platform, and 8 different human movements were identified. It achieves the correct recognition rate of 99.2% on average, which successfully verifies the feasibility and effectiveness of the motion recognition method.
下载PDF全文