基于视觉面部特征疲劳驾驶检测方法  

Fatigue driving detection method based on visual facial features

作　　者：孙翠羽 雷皓安 范谦 马飞虎 谢天长 李明 SUN Cuiyu;LEI Haoan;FAN Qian;MA Feihu;XIE Tianchang;LI Ming(School of Transportation Engineering,East China Jiaotong University,Nanchang 330013,China;Jiangxi Tohui Science and Technology Group Co.,Ltd.,Nanchang 330101,China;Jiangxi Transportation Research Institute Co.,Ltd.,Nanchang 330200,China)

机构地区：[1]华东交通大学交通运输工程学院,南昌330013 [2]江西通慧科技集团股份有限公司,南昌330101 [3]江西省交通科学研究院有限公司,南昌330200

出　　处：《交通科技与经济》2025年第2期57-65,共9页Technology & Economy in Areas of Communications

基　　金：国家重点研发计划项目(2021YFE0105600);江西省03专项及5G项目(20212ABC03W04)。

摘　　要：针对疲劳驾驶引起因素复杂多变、判定标准不明确和难以及时察觉等问题,提出一种基于嵌入式多层感知机(Embedding-MLP)的疲劳驾驶检测方法,驾驶视频数据使用基于RepGhost-Retinaface的人脸面部特征识别模型确定驾驶员人脸位置坐标信息,使用RepGhost-PFLD人脸关键点检测模型提取人脸关键点信息,并将其作为疲劳驾驶识别模型的数据集,提出基于Embedding-MLP的驾驶员疲劳检测模型。结果显示,基于Embedding-MLP的疲劳检测模型AUC值为0.94,Accuracy值为0.87,Precision值为0.87,F1值为0.86,Recall值为0.87,相较于实验中的对比模型,该模型各项评价指标效果最优,有利于对驾驶员疲劳驾驶状态进行识别。Aiming at the problems of complex and variable factors causing fatigue driving,unclear judgment criteria,and difficult to detect in time,we propose an Embedding-MLP(Multilayer Perceptron)based fatigue driving detection method.A fatigue driving detection method based on Embedding-MLP(Multilayer Perceptron)is proposed,which determines the coordinate information of the driver's face position by using the RepGhost-Retinaface face facial feature recognition model based on the driving video data and extracts the face key point information using the RepGhost-PFLD face keypoint detection model,and it is used as the dataset for the fatigue driving recognition model to propose the driver fatigue detection model based on Embedding-MLP.The results show that the fatigue detection model based on Embedding MLP has the AUC value of 0.94,Accuracy value of 0.87,Precision value of 0.87,F 1 value of 0.86 and Recall value of 0.87.Compared with the comparison model in the experiment,the model has the best effect of various evaluation indexes.It is helpful to recognize the driver’s fatigue driving state.

关 键 词：交通工程 疲劳驾驶检测方法 多层感知机 重参数化技术 视觉面部特征 

分 类 号：U419[交通运输工程—道路与铁道工程]