吉林大学学报(信息科学版)

• 论文 • 上一篇    下一篇

基于眼睛与嘴部状态识别的疲劳驾驶检测

邹昕彤, 王世刚, 赵文婷, 赵晓琳, 李天舒   

  1. 吉林大学 通信工程学院, 长春 130012
  • 收稿日期:2016-06-13 出版日期:2017-03-27 发布日期:2017-06-07
  • 作者简介:邹昕彤(1989— ), 女, 吉林松原人, 吉林大学硕士研究生, 主要从事数字图像处理、 疲劳驾驶检测研究, (Tel)86- 13194366163(E-mail)zouxintong822@126. com; 王世刚(1962— ), 男, 长春人, 吉林大学教授, 博士生导师, 主要从事图 像与视频信号智能处理研究, (Tel)86-13504325626(E-mail)wangshigang@ vip. sina. com。
  • 基金资助:
    吉林省科技发展计划基金资助项目(20150204006GX)

Fatigue Driving Detection Based on State Recognition of Eyes and Mouth

ZOU Xintong, WANG Shigang, ZHAO Wenting, ZHAO Xiaolin, LI Tianshu   

  1. College of Communication Engineering, Jilin University, Changchun 130012, China
  • Received:2016-06-13 Online:2017-03-27 Published:2017-06-07

摘要: 为在驾驶员佩戴眼镜的情况下也能准确有效地检测疲劳状态, 提出一种判断是否佩戴眼镜的方法, 并建
立了基于眼睛与嘴部状态的疲劳驾驶检测系统。 对该系统中有关目标检测、 特征提取与图像识别等算法进行
研究。 首先, 采用 Adaboost 算法通过人脸分类器从视频帧中检测人脸区域, 并根据面部器官几何分布规则粗检
眼睛与嘴部区域; 其次, 基于大律法自适应二值化, 采用垂直积分投影法判断是否配戴眼镜, 根据灰度直方图
统计特征值法判断戴眼镜的眼部区域状态, 另外, 利用似圆度判断嘴部打哈欠情况; 最后, 利用 PERCLOS
(Percentage of Eyelid Closure over the Pupil)值识别眼睛疲劳状态, 利用打哈欠频率识别嘴部疲劳状态。 当检测
到驾驶员处于疲劳状态, 则及时给出疲劳警告。 实验结果表明, 该方法可有效解决眼镜对检测的干扰, 并适用
于不同光照与环境。 同时, 在戴眼镜情况下对于眼睛与嘴部疲劳状态的判断优于其他方法。 基本满足疲劳检
测系统对良好的实时性、 稳定性与鲁棒性等要求。

关键词:  疲劳检测, 状态识别, 眼镜判断, 直方图特征

Abstract:  In order to recognize the fatigue state accurately under the condition of the driver wearing glasses, a
method of judging whether to wear glasses is proposed. And the fatigue detection system based on the state of
eyes and mouth is established. Its applied algorithms such as moving object detection, feature extraction and
image recognition and etc are investigated. First, face classifier based on Adaboost algorithm is used to detect
face region from the video frames. The area of eyes and mouth can be detect roughly according to the facial
geometric distribution rules; Secondly, threshold adaptively by Otsu’s method and determine whether driver wear
glasses based on horizontal integral projection method. Then, identify the state of eyes with galsses according to
the method of histogram part statistics characteristic values. In addition, using roundness to judge whether mouth
yawn. Finally, the equivalent PERCLOS(Percentage of Eyelid Closure over the Pupil) value is taken to identify
the fatigue state of eyes, and the frequency of yawning is used to identify the fatigue state of mouth. The system
give an early warning in time when it detects fatigue driving. The experimental results show that these method
can solve the interference of glasses effectively, and is appropriate for different illumination and surroundings.
The accuracy is 95. 8%. It’s better than the method which only use eye or mouth. It can also satisfy the system
requirements of real time, accuracy and robustness.

Key words: glasses judgment, histogram feature,  fatigue detection, state recognition

中图分类号: 

  •