【作者】 盛敬；

【导师】 杨英；

【作者基本信息】 东北大学 ， 车辆工程， 2006， 硕士

【摘要】 目前在频繁发生的交通事故中，驾驶员疲劳已成为引发交通事故的主要因素之一。安全辅助驾驶技术作为智能交通系统的关键技术日益受到人们的关注，其中驾驶员疲劳监控技术对降低交通事故率有着重要的作用。本文研究的驾驶员疲劳监控系统采用机器视觉手段对驾驶员疲劳时嘴部、眼睛的运动状态进行实时监控，机器视觉在实时性、准确性、适用性及经济性等方面比其他监控方法有更大的优势，拓展了疲劳驾驶监控技术的涵盖范围。 论文的研究内容包括：驾驶员面部定位与跟踪、驾驶员嘴唇与眼睛定位、驾驶员嘴唇与眼睛特征向量提取以及驾驶员疲劳状态识别。 根据肤色在颜色空间具有很好的聚类特性的特点，选用YC_bC_r颜色空间作为肤色分布统计的映射空间，建立一个基于肤色的二维高斯分布数学模型，利用基于相似度和人脸形状特性的人脸检测方法，对驾驶员面部进行定位。在驾驶员面部定位的基础上，采用卡尔曼滤波面部跟踪算法进行驾驶员面部跟踪。 嘴唇的红色与面部其它肤色有明显区别，本文提出了一种利用彩色图像中边缘检测和寻找红色像素点相结合的方法对嘴唇区域进行分割和提取。此方法可以有效去除嘴唇周围的阴影干扰和其它干扰，完整地将嘴唇从原图中提取出来。再利用“逼近法”分别对嘴巴处于张开与闭合两种状态时嘴唇的各个特征点进行定位。 驾驶员的双眼精神状态始终保持一致，本文选择右眼为研究对象。根据面部五官的排列规律，以嘴角位置与嘴长为基本点，推算右眼的大概区域。因为眼睑、眼珠都呈天然的黑色，利用图像中的黑色像素点的投影来确定眼睛区域。采用“逼近法”定位右眼睁开与闭合状态时的各个特征点。 在驾驶员疲劳状态识别方面，对BP神经网络进行了修正，提出了一种神经网络群的方法，利用嘴唇、右眼特征向量值作为BP神经网络的输入，利用BP神经网络进行驾驶员疲劳状态识别。这种方法既避免了神经网络的过学习现象导致推广能力下降的缺点，又提高了识别的容错性和准确性。 本文采用vC++6.0开发驾驶员疲劳监控系统中人脸定位与识别的算法，并进行相关的监控试验仿真，以表明上述研究的正确性。更多还原

【Abstract】 Driver’s fatigue factors have been one of the most important causes of traffic accident. Because Safety Driving Assist technology has been the key technology, it has been paid more and more attention to. Driver fatigue monitor is a focus of it. Machine vision has better advantage than other methods on time, accuracy, and adaptability. It makes the scope of Driver fatigue monitor technology wider.The research content of this thesis embraces: the location and tracing of driver’s face, the location of driver’s lip and eye, the extracting character vector of driver’s lip and eye, and the recognition of the driver’ fatigue state.The skin color has the good assembling character in the color room. This thesis selects YC_bC_r color room as distributed and statistical mapping room, and builds a two-dimensional Gause distributed mathematical model through statistics and recognition. It adopts face detecting method based on similarity and the shape of face. Kalman filter is used to track Driver’s face.Using the detecting border method and the searching red color element point method to cut and extract driver’s lip. It can dislodge interference aroud the lip, and pick up the shape of lip completely. Then it uses the approaching method to locate every character point of the lip.In the research of the driver fatigue monitor system, this paper selects the right eye as the object. According to the arrangement of the part of face, it calculates the range of the right eye based on the place of mouth corner and the length of mouth. Because eyelid and eyebal are black, the thesis uses the project of black color element point to locate the domain of the right eye, and the approaching method to locate every character point of the right eye.This paper puts forward a neural network group method, and recognizes driver’s fatigue states by BP neural network group. Experiment results show the BP neural network has fast and steady convergence, and its recognition is accurate and available.This thesis adopts VC++ 6.0 to develop the algorithm of the location and recognization of the driver’s face in the driver fatigue monitor system, and makes the related test and computer simulation to show the correctness of this study.更多还原