【作者】 姜华；

【导师】 崔胜民；

【作者基本信息】 哈尔滨工业大学 ， 车辆工程， 2011， 硕士

【摘要】 在道路交通事故中,驾驶员疲劳驾驶是重要的原因。如果能在驾驶员出现疲劳状态之前就给驾驶员以及时的预警提醒,那么就可以使驾驶员意识到危险并采取相应的措施避免事故的发生,这是汽车主动安全技术的重要方面,所以对驾驶员疲劳状态检测系统的研究具有重要的理论意义与现实意义。本文利用模式识别技术,重点利用其在二类问题分类方面的优势,通过分析驾驶员脸部肤色信息特征,提取驾驶员肤色纹理特征构成特征向量匹配分类器完成驾驶员脸部图像的识别,在此基础上构建准则函数分类器实现眼睛定位识别,从而构建了驾驶员疲劳状态检测系统。本文的核心内容包括驾驶员图像的预处理、驾驶员脸部图像识别、驾驶员脸部追踪、驾驶员眼睛定位,并且选择PERCLOS算法最终实现了疲劳状态检测的预警工作。在驾驶员图像预处理方面利用泛函分析的知识构建了关于真图表面积最小的数学模型实现对驾驶员图像的初步去噪,在此基础上结合形态学腐蚀和膨胀原理完成了驾驶员图像最终去噪工作。为了给后面的眼睛定位工作带来方便,在图像预处理的时候还使用了频域增强、直方图均衡化、傅里叶变换等技术。在驾驶员脸部识别方面提出了利用共生矩阵计算得到的肤色纹理特征构建特征向量匹配分类器的方法将肤色与非肤色进行分类处理,从而实现脸部图像与背景图像的分离;在驾驶员脸部追踪方面利用了camshift算法使得即使在驾驶员头部轻微运动的情况下也能很好的跟踪识别出驾驶员的脸部图像。在驾驶员眼睛定位方面利用纹理特征向量构建了准则函数分类器,并且综合利用了Tanimoto测度的原理最终实现了驾驶员眼睛的识别;最终应用PERCLOS算法完成了驾驶员疲劳状态检测系统的研究。最后,文章对驾驶员疲劳状态检测系统进行了置信度分析,包括识别能力和计算复杂度;还利用ROC曲线对系统进行了有效性验证,结果表明系统基本符合要求。更多还原

【Abstract】 Driver’s fatigue driving is an important reason in the traffic accidents. If we can give the driver warning in time, they can be aware of the dangers and take the according measures to avoid the accidents; this is also an important part of automotive positive safety technology. So, it is very important to study the driver fatigue detection system no matter in theory or practice.This paper takes advantage of pattern recognition especially in dichotomizer to gain the skin vein character based on analyzing driver’s face character and construct the classifier to recognize the face of driver; and then driver’s eye is recognized through functional classifier and pattern recognition technology. At last, the driver fatigue detect system is completed. The centre content of this paper includes image processing, face recognition, tracing face and eye recognition; the PERCLOS algorithm is chosen to realize the driver fatigue detect system.The functional analysis is combined with the morphology to get rid of image noise. During the image processing, enhanced frequency, histogram equalization and Fourier transform are used for convenient of eye recognition.Taking advantage of character vector classifier based on Co-occurrence Matrix Algorithm to distinguish the color of skin and others; and then the face image and the background are separated. Because Camshift algorithm is chosen to realize face tracking, the moving driver face is recognized.The function classifier and the tanimoto measure are used to get the eye of driver. In the end, the driver fatigue detect system is finished by PERCLOS algorithm.At last, the error evaluation is analyzed; including recognition ability and computational complexity, and also the system’s accuracy is checked by ROC curve, the result is satisfying.更多还原