摘要
为减少因疲劳驾驶引发的交通事故,提出融合多参数的驾驶员疲劳检测算法。用渐进校准网络(PCN)检测人脸图像,通过基于CNN的回归模型定位人脸关键点;根据关键点坐标和面部器官的分布规律提取眼睛和嘴部图像,用宽度学习系统(BLS)分别识别眼睛与嘴巴的状态;将眼睛、嘴巴和头部状态的时序序列送入二级宽度网络对司机的状态进行判别。实验结果表明,该算法的疲劳检测准确率为94.9%,单帧检测时间52.43 ms。
To reduce traffic accidents caused by fatigue driving,a multi-parameter driver fatigue detection method was proposed.Human faces were detected through the progressive calibration network(PCN),and face landmarks were located using CNN regression model.According to the distribution of facial organs and landmarks,the images of eyes and mouth were extracted.The states of eyes and mouth were estimated based on broad learning,respectively.A sequence of eye,mouth and head states was fed into a two-level broad learning system to determine the driver’s fatigue state.Experimental results show that the proposed method achieves the detection accuracy of 94.9%with an average processing time of 52.43 ms per frame.
作者
朱玉斌
延向军
申旭奇
卢兆林
ZHU Yu-bin;YAN Xiang-jun;SHEN Xu-qi;LU Zhao-lin(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China;Automation Section,Shanxi Lu’an Group Yuwu Coal Industry Limited Company,Changzhi 046199,China;Changcun Coal Mine,Shanxi Lu’an Environmental Energy Development Limited Company,Changzhi 046102,China)
出处
《计算机工程与设计》
北大核心
2020年第2期537-541,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61771473、61379143、51604271)
江苏省“六大人才高峰”高层次人才基金项目(XYDXX-063)
江苏省“青蓝工程”基金项目(2016)
关键词
疲劳检测
宽度学习
深度学习
信息融合
人脸关键点检测
fatigue detection
broad learning
deep learning
information fusion
facial landmark location