【作者】 刘刚；

【导师】 卢湖川；

【作者基本信息】 大连理工大学 ， 通信与信息系统， 2013， 硕士

【摘要】 随着汽车在国内的大量普及,给人们的工作和生活带来了极大的方便。然而,参差不齐的驾驶水平、淡薄的交通安全意识,导致国内每年发生的交通事故数量居高不下,给国家带来了巨大经济损失,给家庭带来无尽痛苦。经研究发现,在众多的交通事故中,因驾驶员疲劳而导致的事故占相当大的比例。疲劳驾驶已成为交通安全的一大隐患,如何有效检测疲劳驾驶已成为交通安全领域的一个热点问题。本文围绕驾驶员眼睛状态判断,通过检测眼睛睁闭状态和头部姿态判断驾驶员是否处于疲劳状态,并对检测功能进行跨平台应用开发,将检测算法移植到嵌入式终端,把理论算法转化为可实际应用的设备。本文首先介绍了Haar人脸检测分类器的理论基础和训练方法,然后在Haar分类器检测到的人脸上进行眼睛检测。在人眼状态检测方面,经过对各类检测方法的效果和速度进行横向比较,最终选择了基于LBP算法的检测方法。该方法在人眼定位和睁闭状态判断上具有较好的表现,使用样本图像的LBP矩阵检测眼睛位置,再使用匹配数值判断睁闭状态,检测速度快,适合于嵌入式终端运行。在检测到眼睛状态以及头部姿态后,根据各种状态的持续时间判断人员的疲劳状态,发现疲劳状态立刻进行疲劳预警。实验结果表明,基于LBP算法的检测方法对于人脸正面或偏角不大的侧面人脸的疲劳状态具有较好的识别能力,并且对于光线强度的变化具有较好的适应性,经过在实际不同光线环境的汽车内测试,对于疲劳状态的正确识别率达到了90%以上。程序被移植到硬件配置低、运行速度慢的嵌入式终端后,经过对程序多方面速度优化,嵌入式终端的最快处理速度达到了每秒20帧,基本能满足实时性检测要求。本文介绍的检测方法算法复杂度低,检测效果和终端运行速度能够满足实际需求,具有较好的应用前景。更多还原

【Abstract】 The increase of car number has facilitated people’s daily life and work. The number of accidents caused by vehicle every year is very high. It has caused enormous losses for the country and endless sadness for many families. After research, it is found that a mass of accidents were caused by fatigue driving, which has become a hidden danger for traffic safety. So, fatigue driving detection is a hot topics in traffic safety field.This paper centres on eye state research. Detection is performed to find out the eye is whether open or closed, and the face is whether detected or not. According to detection result, we can estimate if the driver were fatigued. Finally, the detection function is transplanted to embedded system, which can be fixed on the car to detect fatigue driving. The algorithm is transformed to practical equipment.First, this paper introduces the theoretical basis and training process of Haar cascade classifier. Then, eye detection is performed in the face area detected by face classifier. After testing many eye detection algorithms, we select the LBP algorithm, which is good at eye location and eye state judgement. Using the LBP matrix of sample image to detect eye position, then, judging eye state according to detection result. The fast detection speed is fit for running in embedded system. After getting eye states, estimating the fatigue state according to the duration of some kind of eye state. If the detection result proves that the driver is tired, detection system will give an alarm. Experimental result shows that this algorithm is good at distinguishing frontal fatigue state, and adapt to illumination intensity changing. After testing in the condition with different illumination intensity, the average recognition rate of this algorithm exceed90percent. When the algorithm is transplanted to the embedded system with low running speed, the fastest average running speed is20frame every second. It can meet the basic needs of real-time detection. Because of these advantages, this detecction method has an extensive application prospect.更多还原