融合DCNN的面部特征检测在驾驶员危险驾驶中的应用研究  

Application of Facial Feature Detection Based on DCNN in Driver s Dangerous Driving

作　　者：王容霞[1] 贺芬[1] 杨伟煌 赵林玲 WANG Rongxia;HE Fen;YANG Weihuang;ZHAO Linling(School of Information Engineering Guangzhou Nanyang Polytechnic College,Guangzhou 510925,China)

机构地区：[1]广州南洋理工职业学院信息工程学院,广东广州510925

出　　处：《商丘职业技术学院学报》2022年第2期71-76,共6页JOURNAL OF SHANGQIU POLYTECHNIC

基　　金：2020年度广东省教育厅高校重点科研项目“基于5G车联网的交通安全预警系统的研究及其应用”(2020ZDZX3096);2019年度广州南洋理工职业学院创新科研团队项目“大数据与智能计算创新科研团队”(NY-2019CQTD-02);2021年度广州南洋理工职业学院校级科研项目“基于深度学习的辅助驾驶安全预警系统的研究”(NY-2021KYYB-02)。

摘　　要：驾驶员疲劳驾驶会增加车辆和行人的交通安全风险.面向BL-DCNN的人脸关键点定位模型,经过精确定位完成嘴巴、眼睛等特征部位的检测.依据定位结果构建融合眼睛闭合时间、眼睛眨眼频率、嘴巴打哈欠状态的疲劳驾驶检测方法,并结合眼睛张角完成眼睛状态的识别.BL-DCNN的人脸关键点检测方法的平均误差和检测时间分别为0.061ms和283ms,检测性能优于其他检测方法.融合多特征的疲劳特征检测方法准确率高达94%.本研究提供了一种能识别驾驶员危险状态的方案,为今后道路交通安全的检测提供了参考.Driver fatigue driving will increase the road safety risk of vehicles and pedestrians.A bl-dcnn oriented face key point location model is proposed.After accurate location,the feature parts such as mouth and eyes are detected.Then,according to the positioning results,a fatigue driving detection method integrating eye closing time,eye blinking frequency and mouth yawning state is constructed,and the eye state recognition is completed combined with eye opening angle.The average error and detection time of bl-dcnn face key point detection method are 0.061 and 283 MS respectively,and the detection performance is better than other detection methods.The accuracy of the multi feature fusion fatigue feature detection method is as high as 94%.The research provides a scheme that can identify the dangerous state of drivers,and provides a reference direction for the detection of road traffic safety in the future.

关 键 词：面部特征 人脸关键点 人脸跟踪 DCNN 疲劳驾驶检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]