自然光照下基于计算视觉的驾驶员疲劳检测研究

【作者】 董文会；

【导师】 吴晓娟；

【作者基本信息】 山东大学 ， 通信与信息系统， 2006， 硕士

【摘要】 随着高速公路的飞速扩展和汽车持有量的迅速增加，公路交通事故明显上升，而驾驶员疲劳驾驶是引发这些交通事故的重要因素之一。据统计，在纽约6.8万起无外界因素引起的严重交通事故中，约10％与疲劳驾驶有关，大约1／4的司机曾有迷糊驾车的经历!在我国每年发生的特大交通事故中，驾驶员疲劳驾驶也是主要的肇事原因。更有研究表明，一个在17～19小时内没有睡眠的司机，其行为能力与一个血液中酒精浓度为0.05％的醉酒司机相当。因此，如何及时且有效的检测出司机的驾驶疲劳以减少事故的发生，成为众多学者研究的重要课题。 20世纪90年代以来，随着计算机和集成电路制造技术的提高，驾驶员疲劳驾驶的研究有了进一步发展，出现了许多有价值的成果。如美国研制的打瞌睡驾驶员侦探系统DDDS(The Drowsy Driver Detection System)、反应时测试仪PVT(ThePsychomotor Vigilance Test)，日本成功研制了电子“清醒带”等，但这些成果都还未能达到实用化。因此，更为实用的疲劳检测方法还在研究中。 本文在深入研究了前人提出的各种疲劳检测方法的原理，分析对比了当前国内外较为流行的各种检测方法并研究了其技术关键和难点后，提出了一种在自然光照下，基于计算机视觉的、能够无接触、无生理负荷、快速有效的驾驶员疲劳检测方法．利用摄像机实时拍摄司机的面部，提取最能反映疲劳程度的眼睛特征，如上眼睑曲率、眼睑距离等，通过模糊融合决策方法，快速、准确地判断甚至预测瞌睡的发生。 本文的主要研究内容： 1、研究了肤色在各颜色空间中的分布特点及各颜色空间与光照的关系，将肤色映射到与光照关系较小的颜色空间中进行肤色分割以消除光照对检测结果的影响。 2、研究并分析了各种快速且有效的人眼定位和跟踪算法，提出了基于规则的，利用脸部二值化后洞对特点定位人眼的方法并通过动态模板匹配实现人眼准确跟踪。 3、研究了眼睛状态特征提取方法，提取出上眼睑曲率和眼睑距离作为判断更多还原

【Abstract】 With the high development of the highway and the quick increase of the number of people having car, the number of traffic accidents are increasing obviously. Driver fatigue problem is one of the important factors that cause traffic accidents. In New York about 10 percent of accidents are fatigue-related in 68 thousand accidents that caused without environment factors and about one fourth of the drivers have ever driven with fatigue. In china, driver fatigue is also the main cause of accidents. It is studied that the behavior of a driver without sleep in 17 to 19 hours equals to a drunk driver whose alcohol concentration is 0.05 percent Therefore, how to supervise and avoid fatigue driving efficiently is one of the significant problems.Since 1990s, with the development of the computer and integrate circuit, the research of driver fatigue has further developed and a lot of valuable products appeared. Such as The Drowsy Driver Detection System and The Psychomotor Vigilance Test in America, The clear-headed strip in Japan. However, those products are not practicable, other more valuable methods are undergoing rigorous evaluation and design.In mis thesis, we investigate and contrast the principles of current methods and technologies for driver fatigue detection, and analyze the key problems and difficulties of these techniques, men propose a quick and effective driver fatigue method based on computer vision in natural illumination. We lay a camera on the panel to shoot the driver’s face. Eye features, such as up-eyelid curvature and the distance of eyelid, which are cues for determining driver’s state, are extracted from the video of face. Then using fuzzy data fusion we can determine even forecast the driver’s state effectively.The contents of this paper are as following:1) Study the skin distribution, the relation of illumination and color space and then segment skin in the color pace which the illumination influences little.2) Study and analyze many eye location and eye tracking methods and propose更多还原