• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:宋佳尧,尉 斌,安姝洁,张涔嘉,杨 玮.基于Dlib的面部疲劳检测模型[J].软件工程,2023,26(12):38-40.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于Dlib的面部疲劳检测模型
宋佳尧, 尉 斌, 安姝洁, 张涔嘉, 杨 玮
(天津商业大学信息工程学院, 天津 300134)
1429633682@qq.com; yubin@tjcu.edu.cn; 2741902510@qq.com; 1249439690@qq.com; 1642087781@qq.com
摘 要: 为了能够准确、实时地对稳定视频情境下人物的打哈欠等疲劳状态进行检测,文章基于Dlib(机器学习的开源库)设计了针对人物面部特征的疲劳状态检测模型。模型包括人脸检测器、关键点定位器、特征计算器和状态预测器四个模块。首先,通过Dlib的相关函数进行人脸检测,然后跟踪定位面部特征点的实时坐标。其次,通过坐标值实时计算与疲劳状态相关度高的PERCLOS(眼睛闭合时间占单位时间的百分率)参数、PMOT(张嘴时间占单位时间的百分率)参数,经对模型训练确定眼部纵横比和嘴部纵横比的阈值,对疲劳状态发出预警提示。经YawDD数据集验证,该模型的平均查全率约为94.2%、平均准确率约为93.3%,能够满足对面部疲劳检测的实时性和准确率的要求。
关键词: 疲劳检测;Dlib;特征点定位
中图分类号: TP181    文献标识码: A
基金项目: 大学生创新创业训练计划项目(202210069107,202310069193)
Facial Fatigue Detection Model Based on Dlib
SONG Jiayao, YU Bin, AN Shujie, ZHANG Cenjia, YANG Wei
(College of In f ormation Engineering, Tianjin University of Commerce, Tianjin 300134, China)
1429633682@qq.com; yubin@tjcu.edu.cn; 2741902510@qq.com; 1249439690@qq.com; 1642087781@qq.com
Abstract: In order to accurately and real-time detect fatigue states of characters such as yawning in stable video scenarios, this paper proposes to design a fatigue state detection model based on Dlib (an open-source library for machine learning) for facial features of characters. The model includes four modules: face detector, key point locator, feature calculator, and state predictor. Firstly, facial detection is performed through the correlation function of the Dlib, and then real-time coordinates of facial feature points are tracked and located. Secondly, the PERCLOS (percentage of eye closure time to a specific time) parameter and PMOT (percentage of mouth opening time to a specific time) parameter, which are highly correlated with the fatigue state, are calculated in real time through the coordinate values. After training the model, the threshold values of the eye aspect ratio and mouth aspect ratio are determined, so as to issue an early warning prompt for the fatigue state. Validated by the YawDD dataset, the model achieves an average check rate of about 94.2% and an average accuracy of about 93.3% , which can meet the requirements of real-time and accuracy for facial fatigue detection.
Keywords: fatigue detection; Dlib; feature point localization


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫