【作者】 王芳；

【导师】 胡涛；

【作者基本信息】 西安理工大学 ， 信号与信息处理， 2009， 硕士

【摘要】 由疲劳驾驶引发的交通事故已经导致了国家经济和人民生命财产的巨大损失,因此急需研究出一种实时、有效的驾驶员疲劳状态检测方法。本文对驾驶员疲劳状态检测的研究现状进行了详细的了解后,认为基于眼睛状态的疲劳状态检测方法具有最好的发展前景,并对其进行了深入的研究。首先,本文选用基于AdaBoost的方法进行人脸检测和眼睛状态判断,主要对基于AdaBoost的眼睛状态判断方法进行了详细的研究,训练出AdaBoost眼睛状态分类器,并进行了详细的测试。然后基于PVT (psychomotor vigilance test)测试,本文组织建立了模拟驾驶视频库,并依据PVT测试得分对每段视频的疲劳程度进行了等级量化,为评价本文疲劳状态判断方法的可行性提供了依据。本文对部分视频进行处理,提取每帧的眼睛疲劳参数：PerClose、闭眼持续时间(TClose)和眨眼频率(ECF),并对他们与疲劳程度的相关性进行了研究,最后采用自适应模糊神经网络的方法来得到模糊推理系统,并依据此模糊推理系统对驾驶员的疲劳状态进行判断。实验结果表明,本文的方法耗时小于135毫秒,能够实时、有效的判断驾驶员的眼睛状态,并提取眼睛疲劳参数,最后对驾驶员的疲劳状态做出较为准确的判断。更多还原

【Abstract】 Traffic accidents caused by driving in case of fatigue have led to huge losses of the national economy and people’s lives, so there is an urgent need to develop a real-time and valid driver’s fatigue detection method.After a detailed understanding of the actuality of the driver’s fatigue detection research, this paper thinks that the fatigue detection method based on the eye states has the best prospects, and then studies it in-depth.First of all, the paper uses the method based on AdaBoost on detecting the face and judging the state of eyes. It carries out a detailed research on the eyes state estimation method based on AdaBoost, and trains an AdaBoost classifier for eye state estimation. Then, the paper sets up a simulated driving video library based on the PVT (psychomotor vigilance test) test, and quantifies the fatigue degree of each video based on the PVT score. The degree quantification provides a basis for evaluating the feasibility of the driver’s fatigue state judgement method. After some video was treated, the fatigue parameters of eyes:PerClose, the closing duration of eyes (TClose) and blinking frequency (ECF) are extracted. The paper studies the correlation of the fatigue parameters of eyes and the fatigue degree. Finally, the paper uses the adaptive fuzzy neural network to get a fuzzy inference system, and uses the fuzzy inference system to judge the driver’s fatigue state in experiments.The experimental results show that this method’s management time is less than 135 milliseconds. This method can effectively estimate the driver’s eyes state in real-time, extract the fatigue parameters of eyes, and judge the driver’s fatigue state finally.更多还原