【作者】 魏小辉；

【导师】 胡正平；

【作者基本信息】 燕山大学 ， 通信与信息系统， 2010， 硕士

【摘要】 随着交通事业的迅猛发展,交通安全已成为全球范围内普遍关注的重要问题。然而,每年因交通事故丧失生命的人数依然居高不下,各类恶性交通事故仍然接连发生,经研究发现,其中因驾驶员疲劳驾驶造成的事故占相当大的比例,由此可见,疲劳驾驶已成为伤害人身安全的一大隐患。因而,驾驶员疲劳状态检测研究及预警系统设计成为近几年国内外研究的热点。针对该问题,本文从驾驶员面部特征入手,通过识别眼部及嘴唇的状态判断驾驶员是否发生疲劳,并给予及时预警。本文主要工作如下:首先,讨论并研究了一种多姿态环境下基于局部Gabor方向投影子空间的多状态模板人眼定位方法。该方法首先采用AdaBoost人脸检测算法从输入的彩色图像序列中得到人脸区域,并通过光照补偿对人脸进行预处理,以消除光照不均对实验效果的影响,再依据人脸结构特征,对人脸头部姿势进行估计和校正,最后,利用多尺度水平方向Gabor变换后的人脸图像在眉眼处幅值较大的特点,通过局部投影方法得到左右眉眼的大致区域,并在此区域中采用多状态模板匹配进行人眼精确定位。其次,在对传统Mean Shift跟踪算法分析研究的基础上,对其进行改进,采用人眼颜色特征与局部二值模式(LBP)纹理特征相结合的表示方法对人眼进行描述,并将其成功嵌入Mean Shift算法实现人眼跟踪。最后,利用Canny算子对人眼进行边缘提取,通过测量上下眼睑距离进行眨眼检测,再利用唇色信息在人脸图像中找出嘴唇部位,根据嘴唇宽高比描述嘴唇的张开程度,通过打哈欠检测与PERCLOS原理相结合判断驾驶员是否发生疲劳。经实验验证,本文提出的方法有效可行,对不同光照及人脸姿态下的疲劳状态实时检测及跟踪具有较高的鲁棒性。更多还原

【Abstract】 With the rapid development of transport, the traffic safety has become an important issue be commonly concerned on a global scale. However, each year the figure of people killed in traffic accidents is remaining obstinately high, and all kinds of vicious traffic accidents are still occur. Study shows that most of them are caused by the drivers’fatigue, so driver fatigue has become a hazard to personal safety. Therefore, the study and design of safety assist systems specific to driver fatigue driving attracted much attention of researchers, and become a hotspot all over the world.For the problem, the paper select the facial features of driver as the judgement method to detect the driver fatigue and give timely warning. The main work done in this paper is as follows:Firstly, a method of eye localization and detection based on multimode template matching in projection subspace of local gabor derection under multi-pose condition is presented. The face region is obtained from input color image by AdaBoost algorithm of face detection firstly; Secondly,lighting compensation in face region is done to eliminate the effects of illumination,then the lip is detected according to the lips color, next the left and right corners are detected by harris corner detection algorithm, and the perpendicular bisector of the corners is the symmetry of the face, then the deflection angle of the face is estimated. Finally, gabor transform is used in the corrected face, and local gray projection is performed to locate the rough areas of eyes, then using the multimode template matching in the areas to locate the eyes and judge the state of eyes.Secondly, by analyzing the traditional Mean Shift algorithm, a new approach based on LBP descriptor embedded Mean Shift algorithm for eye and mouth fatigue state tracking is presented in the thesis. The texture features of local binary pattern (LBP) and the color features are combined with together effectively to describe the human eye and embeded Mean Shift algorithm for realizing eye tracking.Finally, the edge image is got by Canny operator from the original image of eye, and then we detecte the blink by measuring the distance of the upper and lower eyelids, next the lip is detected according to the lips color, and the state of mouth is described by the aspect ratio of the lip. In order to make the algorithm more valid, we combined PERCLOS with yawning detection to detect driver’s drowsy state, if yawning is detected at the same time, the driver will be judged to fatigue. Experimental results show that the proposed algorithm is effective and feasible, and robust to variations of background, illumination and pose.更多还原