东北大学学报:自然科学版 ›› 2019, Vol. 40 ›› Issue (7): 937-941.DOI: 10.12068/j.issn.1005-3026.2019.07.005

• 信息与控制 • 上一篇    下一篇

基于短时心电信号的疲劳驾驶检测算法

徐礼胜1,2, 张闻勖1, 庞宇轩3, 吴承暘1   

  1. (1. 东北大学 中荷生物医学与信息工程学院, 辽宁 沈阳110169; 2. 沈阳东软智能医疗科技研究院有限公司, 辽宁 沈阳110167; 3. 东北大学 计算机科学与工程学院, 辽宁 沈阳110169)
  • 收稿日期:2018-07-02 修回日期:2018-07-02 出版日期:2019-07-15 发布日期:2019-07-16
  • 通讯作者: 徐礼胜
  • 作者简介:徐礼胜 (1975-),男,安徽安庆人,东北大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(61773110,61374015); 沈阳东软智能医疗科技研究院有限公司开放课题基金资助项目(NRIHTOP1801); 中央高校基本科研业务费专项资金资助项目(N161904002).

Driver Drowsiness Detection Algorithm Using Short-Time ECG Signals

XU Li-sheng1,2, ZHANG Wen-xu1, PANG Yu-xuan3, WU Cheng-yang1   

  1. 1.School of Sino-Dutch Biomedical and Information Engineering, Northeastern University, Shenyang 110169, China; 2. Neusoft Research of Intelligent Healthcare Technology Co., Ltd., Shenyang 110167, China; 3. School of Computer Science & Engineering, Northeastern University, Shenyang 110169, China.
  • Received:2018-07-02 Revised:2018-07-02 Online:2019-07-15 Published:2019-07-16
  • Contact: XU Li-sheng
  • About author:-
  • Supported by:
    -

摘要: 心率变异性分析是最常用的一种基于心电信号的疲劳驾驶检测方法.然而,该方法需要被检测信号时间足够长,且准确率较低.因此提出一种基于短时心电信号的疲劳驾驶检测算法.首先,按照30s的时长截取短时心电信号序列,利用差分阈值法确定R波位置,根据R-R间期差值大小剔除不合格的噪声样本;然后,计算R-R间期序列的时域/频域特征并与利用ImageNet数据集预训练的深度卷积神经网络模型提取的特征相结合;最后,设计了一种随机森林分类器并基于这些特征进行分类.结果表明,该算法在疲劳驾驶检测上具有良好的分类效果,平均准确率达到91%.因此,相较于心率变异性分析方法,本算法检测所需心电信号更短,且在准确率上具备显著优势.

关键词: 心电信号, 疲劳驾驶, 随机森林, 迁移学习, 神经网络

Abstract: Heart rate variability analysis is used extensively for detecting driver drowsiness based on ECG signals. However, this method is deficient in accuracy and needs long-time ECG signal. An algorithm for driver drowsiness detection based on short-time ECG signals was proposed. First, the original ECG signal is rearranged into 30s segments, after which the R-wave positions are extracted using differential threshold algorithm and the noisy segments are excluded according to the calculated R-R interval. Then, time and frequency domains’ features of R-R interval series were extracted and combined with the features obtained by the deep convolutional neural network model with pre-trained weights of ImageNet dataset. Finally, random forest classifier was employed to detect the fatigue status of drivers based on the extracted features. The results demonstrate that the proposed algorithm has good performance in detecting driver drowsiness, with an averaged overall accuracy of 91%. The proposed algorithm needs shorter ECG signals and has higher accuracy in detecting driver drowsiness.

Key words: ECG signal, driver drowsiness, random forest, transfer learning, neural network

中图分类号: