| 摘 要: 空巢老人跌倒受伤，如不及时就医，可能会导致终身残疾，甚至危及生命。为此设计基于YOLOv5(单阶段目标检测算法)的空巢老人跌倒检测系统。首先自建跌倒数据集约7,000 张，按1:4划分为训练集和验证集训练；然后搭建监控视频流、编写可视化检测客户端、部署服务器端，同时将短信阈值及进程融合到检测系统，实现老人跌倒行为的及时反馈；最后对约100 个实际监控视频进行测试、分析。实验结果表明，系统对跌倒和正常两种状态的检测平均精度为94.23%，跌倒后正确发送短信为45 次，发送成功率约98.10%，能达到较理想的效果。 |
| 关键词: 深度学习；YOLOv5；跌倒检测 |
|
中图分类号: TP391.4
文献标识码: A
|
| 基金项目: 广西民族大学中国-东盟研究中心创新研究团队课题(TD201405)；广西重点研发计划项目(AB16380199)；广西高校中青年教师基础能力提升项目(KY2016YF011). |
|
| Design and Research of Fall Detection System for Empty-nesters based on YOLOv5 |
|
XU Wei, LIAO Yikui
|
(College of Electronic Information, Guangxi University for Nationalities, Nanning 530006, China)
1293416157@qq.com; 402131612@qq.com
|
| Abstract: If an empty-nester falls and gets injured, failure to be taken to the hospital in time may result in lifelong disability and even life-threatening injuries. Aiming at the problem, this paper proposes to design a fall detection system for empty-nesters based on YOLOv5 (a single stage target detection algorithm). The specific research scheme is as follows: firstly, about 7,000 fall data sets are built and divided into training set and verification set with the ratio of 1:4. Then, the monitoring video stream is built, the visual detection client is designed, and the server is deployed. At the same time, SMS threshold and SMS process are integrated into the detection system to realize the timely feedback of the elderly falling behavior. Finally, more than 100 actual monitoring videos are tested and analyzed. The experimental results show that the average detection accuracy of the system for falling and normal states is 94.23%; the correct sending of short messages after falling is 45 times, and the sending success rate is about 98.10%, which can achieve ideal results. |
| Keywords: deep learning; YOLOv5; fall detection |
