基于脑电与面部特征拼接融合的驾驶疲劳识别研究  

Study on Driving Fatigue Recognition Based on Splicing andFusion of EEG and Facial Features

作　　者：郭寒英 王诗麟 刘双侨 董文安 卓小军 王星捷 唐立 乔少杰 GUO Hanying;WANG Shilin;LIU Shuangqiao;DONG Wenan;ZHUO Xiaojun;WANG Xingjie;TANG Li;QIAO Shaojie(School of Automobile and Transportation,Xihua University,Chengdu 610039,China;Sichuan Easy Simulation Smart Technolog Co.Ltd.,Chengdu 610094,China;Chengdu Jiaotou Information Technology Co.Ltd.,Chengdu 610073,China;Sichuan Joomon Sci-Tech Co.Ltd.,Chengdu 610095,China;School of Computer Science and Technology,Yibin University,Yibin,644000,China;School of Software Engineering,Chengdu University of Information Technology,Chengdu 610225,China)

机构地区：[1]西华大学汽车与交通学院,四川成都610039 [2]四川易方智慧科技有限公司,四川成都610094 [3]成都交投信息科技有限公司,四川成都610073 [4]四川九门科技股份有限公司,四川成都610095 [5]宜宾学院计算机科学与技术学院,四川宜宾644000 [6]成都信息工程大学软件工程学院,四川成都610225

出　　处：《无线电工程》2024年第12期2820-2830,共11页Radio Engineering

基　　金：四川省科技计划项目(2023NSFSC0386)。

摘　　要：使用面部特征和脑电(Electroencephalogram, EEG)特征识别驾驶员的疲劳状态,对驾驶员进行疲劳提醒,可以有效降低事故发生概率。为解决单一面部特征或EEG特征识别精度不高的问题,提出一种基于EEG与面部特征拼接融合的疲劳识别方法。提取EEG信号的时域、频域、非线性特征和面部特征,通过特征层信息融合的方法进行特征拼接。为提高面部特征识别速度,提出了一种改进的YOLOv5_mobilenet模型。在此基础上,将拼接后的融合特征通过六大机器学习模型进行精度识别,并选择准确性、F1_score、精确率和召回率进行分析、评价。使用公开的数据集来验证所提出的方法,结果表明,改进的YOLOv5_mobilenet模型在各个特征表现均高于现有模型;不同的机器学习模型评价结果显示,与单一的疲劳特征识别相比融合特征表现更好,因此,基于EEG与面部特征拼接的融合特征用于驾驶疲劳识别是可行的。Using facial and Electroencephalogram(EEG)features to identify the driver's fatigue state and provide fatigue reminders can effectively reduce the probability of accidents.To solve the problem of low recognition accuracy of single facial features or EEG features,a fatigue recognition method based on the fusion of EEG and facial features is proposed. Firstly, the time-domain, frequency-domain, nonlinear features, and facial features ofthe EEG signal are extracted, and feature concatenation is performed through feature layer information fusion. Toimprove the speed of facial feature recognition, an improved YOLOv5_mobilene model is proposed. On this basis,the fused features after splicing are identified for accuracy through six major machine learning models, and accuracy,F1_store, precision, and recall are selected for analysis and evaluation. The proposed method was validated using apublicly available dataset, and the results showed that the improved YOLOv5_mobilene model outperformedexisting models in all feature performances. The evaluation results of different machine learning models show thatfused features perform better than single fatigue feature recognition. Therefore, it is feasible to use fused featuresbased on EEG and facial feature concatenation for driving fatigue recognition.

关 键 词：脑电信号特征 面部图像特征 特征融合 疲劳识别 机器学习 

分 类 号：U491.2[交通运输工程—交通运输规划与管理]