基于OpenPose改进的老人摔倒检测算法  

Improved Elderly Fall Detection Algorithm Based On OpenPose

作　　者：胡昕 刘瑞安[1] 黄玉兰 HU Xin;LIU Ruian;HUANG Yulan(College of Electronics and Communication Engineering,Tianjin Normal University,Tianjin 300387,China)

机构地区：[1]天津师范大学电子与通信工程学院,天津300387

出　　处：《现代信息科技》2023年第23期73-78,82,共7页Modern Information Technology

基　　金：天津师范大学研究生科研创新项目(2022KYCX105Y)。

摘　　要：为避免老人摔倒后未能及时提供医疗援助而造成的人身伤害,研究发现老人摔倒并及时发出警告,可减少老人摔倒的损失和严重后果。为了提高老人摔倒检测算法的检测精度和实时性能,提出了一种基于OpenPose改进的老人摔倒检测算法。该算法在OpenPose人体骨架信息识别网络的基础上,提出将其部分卷积层替换为深度可分离卷积神经网络类型。该算法使用长短期记忆神经网络来检测老人的摔倒。从URFall公共数据集提取跌倒和相关行为数据,丰富自制数据集,实验结果表明,本文改进后算法大大提升了系统判别摔倒的识别精度。In order to avoid personal injuries caused by failure to provide timely medical assistance after an elderly fall,the study finds that elderly falls and timely warnings can reduce the damage and serious consequences of elderly falls.In order to improve the detection accuracy and real-time performance of the elderly fall detection algorithm,an improved elderly fall detection algorithm based on OpenPose is proposed.The algorithm proposes to replace some of its convolutional layers with the Depthwise Separable Convolution neural network type based on the OpenPose human skeleton information recognition network.The algorithm uses the Long Short-Term Memory Networks to detect falls of the elderly.The fall and related behavioural data are extracted from the URFall public dataset to enrich the home-made datasets.Experimental results show that the improved algorithm in this paper greatly improves the recognition accuracy of the system in discriminating falls.

关 键 词：OpenPose 深度可分离卷积 长短期记忆神经网络 摔倒检测 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]