作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (8): 243-250. doi: 10.19678/j.issn.1000-3428.0059421

• 图形图像处理 • 上一篇    下一篇

融合人眼特征与深度学习的疲劳驾驶检测模型

樊星1, 刘占文2, 林杉1, 窦瑞娟2   

  1. 1. 长安大学 电子与控制工程学院, 西安 710064;
    2. 长安大学 信息工程学院, 西安 710064
  • 收稿日期:2020-09-02 修回日期:2020-11-11 发布日期:2020-11-16
  • 作者简介:樊星(1994-),女,硕士,主研方向为计算机视觉、状态识别、深度学习;刘占文,副教授、博士;林杉,讲师、博士;窦瑞娟,硕士研究生。
  • 基金资助:
    国家自然科学基金(61703054);陕西省重点研发计划(2018ZDXM-GY-044)。

Fatigue Driving Detection Model Fusing with Human Eye Features and Deep Learning

FAN Xing1, LIU Zhanwen2, LIN Shan1, DOU Ruijuan2   

  1. 1. School of Electronics and Control Engineering, Chang'an University, Xi'an 710064, China;
    2. School of Information Engineering, Chang'an University, Xi'an 710064, China
  • Received:2020-09-02 Revised:2020-11-11 Published:2020-11-16

摘要: 针对现有疲劳驾驶检测技术不能有效平衡准确性和实时性的问题,通过融合人眼特征与深度学习,构建一种新的疲劳驾驶检测模型。设计GP-VGG16网络进行眼部状态识别,通过将人工先验信息集成到轻量级深度网络中,提高眼部状态识别的准确性、稳定性和实时性。在此基础上,利用眼部特征-疲劳等级模型将疲劳状态划分为9个等级,定量估计驾驶员状态,同时基于少样本学习建立高效的自动标签生成网络,减少对大量无标签驾驶数据的语义标注。实验结果表明,该模型的准确率达到97.1%,运行速度达到39.96 frame/s,能够有效提高驾驶员疲劳状态识别的准确性与时效性。

关键词: 疲劳驾驶检测, 深度学习, 少样本学习, GP-VGG16网络, 积分投影

Abstract: To address the problem that the existing fatigue driving detection technology cannot balance the accuracy and real-time performance effectively, a fatigue driving detection model based on human eye features and deep learning is proposed. GP-VGG16 network is designed for eye state recognition. By integrating artificial prior information into lightweight deep network, the accuracy, stability and real-time performance of eye state recognition are effectively improved. On this basis, the eye feature-fatigue level model divides the fatigue state into nine grades, and quantitatively estimates the driver's state. At the same time, an efficient automatic label generation network is established based on few-shot learning to reduce the semantic annotation of a large number of unlabeled driving data. The experimental results show that the accuracy of this method is 97.1%, and the running speed is 39.96 frame/s. The method effectively improves the accuracy and real-time performance of driver fatigue state recognition.

Key words: fatigue driving detection, deep learning, few-shot learning, GP-VGG16 network, integral projection

中图分类号: