基于深度学习的毫米波雷达人体摔倒检测系统研究

Research on millimeter wave radar human fall detection system based on deep learning

针对现有摔倒检测系统难以完成全天时检测、存在侵犯被检测人隐私的问题,该文设计了一种基于深度学习的毫米波雷达人体摔倒检测系统,包括信号采集、训练数据生成、智能检测和显示与告警四个部分。该系统利用1642毫米波雷达采集数据,对数据进行短时傅里叶变换,经数据增强后构建时频图数据集,通过ResNet101网络进行动作检测。检测为摔倒动作后,向远程接收端发送报警信息。该系统能够检测摔倒、弯腰、下蹲三种动作。实测结果表明,检测准确率为94.3%。

Aiming at the problem that the existing fall detection system is difficult to complete all-day detection and violates the privacy of the detected person,a deep learning⁃based millimeter wave radar human fall detection system is designed,which includes four parts:signal acquisition,training data generation,intelligent detection and alarm display.The system uses 1642 millimeter wave radar to collect data,and performs short⁃time Fourier transform on the data.After data enhancement,the time⁃frequency picture dataset is constructed,and the action detection is carried out through ResNet101 network.If the detection result is fall,alarm information will be sent to the remote receiver.The system can detect three movements including falling,bending and squat.The measured results show that the accuracy of detection is 94.3%.