节点文献

基于深度融合残差网络的驾驶员眼睛状态检测

Driver eye state detection based on deep fusion residual network

  • 推荐 CAJ下载
  • PDF下载
  • 不支持迅雷等下载工具,请取消加速工具后下载。

【作者】 王国栋王增才范佳城

【Author】 Wang Guodong;Wang Zengcai;Fan Jiacheng;School of Mechanical Engineering,Shandong University;Key Laboratory of High-Efficiency and Clean Mechanical Manufacture,Ministry of Education,Shandong University;National Demonstration Center for Experimental Mechanical Engineering Education,Shandong University;

【通讯作者】 王增才;

【机构】 山东大学机械工程学院山东大学高效洁净机械制造实验室教育部重点实验室山东大学机械基础实验教学中心国家级实验教学示范中心

【摘要】 驾驶员眼睛状态检测是驾驶员疲劳检测的重要组成部分。为有效解决实际驾驶环境中驾驶员眼睛状态检测问题,提出了一种基于深度融合残差网络的方法。该方法将深度神经网络与深度卷积神经网络相融合,利用深度神经网络对驾驶员眼睛特征进行识别,利用深度卷积神经网络对驾驶员眼睛图像进行分析,最终根据二者检测结果的加权平均值对实际驾驶环境下驾驶员眼睛状态做出判定。模型中深度卷积神经网络部分在多通道卷积的基础上,结合了残差网络和深度模型压缩策略,提升眼睛状态检测精度的同时提高了检测速度。相关实验结果表明,该方法在实验环境和实际环境下与其他已有的方法相比检测精度更高、计算速度更快。

【Abstract】 Driver eye state detection is an important part of driver fatigue detection. To effectively solve the problem of driver’s eye state detection in the actual driving environment, a method based on deep fusion residual network is proposed. The model combines a deep neural network with a deep convolutional neural network, which uses deep neural network to identify driver’s eye features and uses deep convolutional neural network to analyze driver’s eye image. Finally, the driver’s eye state in the actual driving environment is determined based on the weighted average of the detection results of the two parts. On the basis of multi-channel convolution, the deep convolutional neural network in the model combines residual network and deep model compression strategies to improve the accuracy and the speed of eye state detection. The experimental results show that the model has higher detection accuracy and faster calculation speed than other existing methods both in the experimental environment and the actual environment.

【基金】 山东省自然科学基金资助项目(ZR2018MEE015)
  • 【文献出处】 机械设计与制造工程 ,Machine Design and Manufacturing Engineering , 编辑部邮箱 ,2021年09期
  • 【分类号】U463.6;U492.8;TP391.41;TP183
  • 【被引频次】2
  • 【下载频次】94
节点文献中: 

本文链接的文献网络图示:

本文的引文网络