基于动态匹配模型的驾驶人嘴部行为识别与分级预警  

Driver Mouth Behavior Recognition and Hierarchical Prewarning Based on Dynamic Matching Model

作　　者：付锐[1] 程文冬[2,3] 张名芳[2] 袁伟[2] 刘卓凡[2] 郭艳君[2] 

机构地区：[1]长安大学,汽车运输保障技术交通行业重点实验室,西安710064 [2]长安大学汽车学院,西安710064 [3]西安工业大学机电工程学院,西安710032

出　　处：《汽车工程》2015年第9期1095-1102,共8页Automotive Engineering

基　　金：长江学者和创新团队发展计划项目(IRT1286);国家自然科学基金(61374196和61473046);中央高校基本科研业务费专项资金(2013G2221006,2013G1502060和2013G1502062)资助

摘　　要：鉴于嘴部行为能够反映出驾驶人不同的精神状态,本文中提出了一种基于机器视觉的嘴部行为识别与精神状态预警方法。首先采用改进的高斯混合模型进行光照自适应的唇色分割。然后在嘴唇区域内建立了基于灰度能量角点的嘴唇动态匹配模型,用于描述嘴部特征并提取动作参数。最终建立了嘴部行为的判别模式,实现了驾驶人精神状态的分级预警。实验结果表明,动态匹配模型具有较高的嘴部特征匹配精度和动作适应能力。算法能够有效检测"闭嘴"、"说话"和"打哈欠"行为,分心预警和疲劳预警的监测精度分别为85.8%和92.9%。In view of that mouth behaviors can reflect different mental states of driver, a machine visionbased mouth behavior recognition and mental state prewarning method for driver is proposed in this paper. Firstly an improved Gaussian mixed model is adopted for light-adaptive tip segmentation. Then a dynamic matching model for lip region is proposed based on gray-scale energy corners to describe lip features and extract mouth movement parameters. Finally a discrimination pattern of mouth behavior is built to fulfill the hierarchical prewarning of driver＇s mental states. Experimental results demonstrate that the dynamic matching model proposed has high matching accuracy in mouth features and strong adaptability to mouth movement. The algorithm can effectively detect the mouth behaviors of ＂closed＂, ＂speaking＂ and ＂yawning＂. The monitoring accuracies of distraction and fatigue prewarning reach 85.8% and 92. 9% respectively.

关 键 词：嘴部行为 高斯混合模型 能量角点 动态匹配模型 分级预警 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]