摘要
复杂的交通环境、个人和社会因素制约了疲劳驾驶识别技术的应用效果,提出一种对视频中驾驶员脸部状态和车辆驾驶状态数据进行融合分析的疲劳驾驶识别算法。该算法基于Dlib库提取的人脸轮廓点计算眼和嘴的纵横比值,生成眯眼和哈欠特征,基于线性拟合趋势提取法生成车辆操控活跃度特征,然后采用改进后的随机森林模型对疲劳状态进行识别。该模型基于权重对特征的重要性进行评估,提高了树节点分裂的有效性,并给出了森林中树的数量的调控方法。实验结果表明所提算法的疲劳驾驶识别准确率均值达到了92.06%,并具有较好的计算效率,验证了其有效性。
Complex traffic environment,personal and social factors restrict the application effect of fatigue driving recognition technology.This paper presents a fatigue driving recognition algorithm based on the fusion analysis of driver’s face state in video and vehicle driving state data.The algorithm calculates the aspect ratio of eyes and mouth based on the extracted face contour points using Dlib database,and then generates the orbital and yawn features.At the same time,the vehicle manipulation activity features based on the linear fitting trend extraction method are obtained.The improved random forest model is used to identify the fatigue state.The model evaluates the importance of features based on weight,improves the validity of tree nodes splitting,and gives the method of regulating the number of trees in forest.The experimental results show that the average accuracy of fatigue driving recognition of the proposed algorithm reaches 92.06%,and it has good computational efficiency meanwhile,which verifies its effectiveness.
作者
吴士力
唐振民
刘永
WU Shili;TANG Zhenmin;LIU Yong(School of Computer Science and Engineering,Nanjing University of Technology,Nanjing 210094,China;Laboratory of Chang’an Ford,Department of Automobile Engineering,Nanjing Vocational Institute of Transport Technology,Nanjing 211188,China)
出处
《计算机工程与应用》
CSCD
北大核心
2020年第20期212-219,共8页
Computer Engineering and Applications
基金
国家自然科学基金(No.61305134)。
关键词
随机森林
人脸轮廓点
车辆操控活跃度
疲劳驾驶
random forest
face contour points
vehicle manipulation activity
fatigue driving
