基于YOLO目标检测的摔倒报警系统  

Fall Detection and Alarm System Based on YOLO Target Detection

作　　者：秦梓竣 邓君[1] 陈坤豪 王岩 Qin Zijun;Deng Jun;Chen Kunhao;Wang Yan(School of Mechanical Engineering,Dongguan University of Technology,Dongguan Guangdong 523000,China)

机构地区：[1]东莞理工学院机械工程学院,广东东莞523000

出　　处：《机电工程技术》2024年第1期224-227,共4页Mechanical & Electrical Engineering Technology

摘　　要：针对人们摔倒后可能出现意识模糊无法呼救易产生二次伤害的问题,研究开发了一款可通过实时检测摄像头画面内容人物是否摔倒,并在发现摔倒后及时反馈警报的系统。通过机器视觉技术结合深度学习方法,首先自建数据集并加以训练得到用于构建基于YOLOv5的人物摔倒状况的模型,然后在此基础上结合实时图像传输与下位机Arduino板进行控制响应,实现对人物摔倒状况的快速检测和摔倒后的警报响应,通过对比YOLOv5m和YOLOv5s的loss速度对比确定了训练模型。实验结果表明,所设计的摔倒检测报警系统模型训练精确率为99.52%,召回率为99.77%,可以分辨出侧摔,仰摔等摔倒情况,实际测试结果置信度在0.6~0.8,能够发现摔倒情况并发出警报,提高了救助摔倒者的可能性。In order to solve the problem that people may become unconscious and unable to call for help after falling,which may easily cause secondary injuries,a system is developed that can detect whether the person in the camera screen has fallen in real time,and provide timely feedback alerts when a fall is detected.By combining machine vision technology with deep learning methods,a data set is built and trained to obtain a model for building a falling situation of a figure based on YOLOv5.Then,on this basis,the real-time image transmission and the Arduino board of the lower machine are combined to control the response,so as to achieve rapid detection of the fall condition of the figure and the alarm response after the falling.The training model is determined by comparing the loss speed of YOLOv5m and YOLOv5s.Experimental results show that the designed fall detection and alarm system has a model training accuracy of 99.52%and a recall rate of 99.77%.It can distinguish side falls,backward falls and other fall situations,and the actual test result confidence level is 0.6~0.8.The system can detect a fall and sound an alarm,improving the possibility of rescuing the person who fell.

关 键 词：摔倒 报警 目标检测 深度学习 YOLOv5 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]