【作者】 王秀；

【导师】 朱学峰；

【作者基本信息】 华南理工大学 ， 模式识别与智能系统， 2010， 硕士

【摘要】 随着交通运输业的快速发展,交通事故已经成为各个国家面临的难题之一。统计分析表明,驾驶员疲劳为导致交通事故的重要原因之一。随着计算机的快速发展和应用,采用摄像头采集到的驾驶员头部图像通过各种算法来判断驾驶员的疲劳状态并加以提醒已经成为了现在疲劳检测的主流方向。本文主要思想是通过对驾驶员脸部信息的获取和图像检测,进行疲劳判断。脸部信息的获取包括进行脸部位置的检测,眼睛的检测定位和嘴唇的检测定位。脸部位置的检测定位主要建立肤色模型以及计算肤色相似度图,通过阈值分割方法进行分类和定位。在得到了脸部位置的准确信息后,进行眼睛和嘴唇的检测。根据人的面部特征首先对检测目标进行粗略定位,然后通过边界提取的办法进行眼睛的定位和状态的计算以及利用Fisher线性分类器进行嘴唇的定位和状态的计算。在得到了脸部信息的状态值后,进行驾驶员疲劳程度的综合评判,制作模糊控制器定义输入输出以及模糊推理规则。本文对驾驶员疲劳的检测进行了系统的研究与设计,并通过实验测试其准确性和有效性。本文的主要研究工作主要分为以下几个部分:第一,改善了传统的边缘检测定位眼睛位置噪声多,错误率高的方法,提出边界提取方法进行定位,实验证明此方法是快速和高准确率的。改善了由于受到外界诸多因素的影响而检测效果不好的情况。第二,运用Kalman滤波方法实现了对跟踪目标物体的跟踪和对噪声的滤波,加快了检测速度,进一步的提高了检测的准确性。第三,提出综合评判驾驶员疲劳程度的概念,利用模糊控制器制定控制规则对疲劳状态进行综合评判,给检测留有一定容错率。更多还原

【Abstract】 With the rapid development of transport, traffic accidents have become one of the various problem our country is facing. Statistical analysis showed that driver fatigue is one of the important reasons which is leading to traffic accidents. With the rapid development and application of computer, using camera to collected images of the driver and to determine the fatigue level of the driver by many algorithm.The main idea of this paper is to determine the fatigue level by getting the information of the driver face and image detecting. The information of driver face include detect and calculate the statement of the eye ,mouth head position. The method of face detection is establishment of skin color model and computation of similarity graph, then classify by threshold segmentation. After get the accurate head position, detecting the eye and mouth position. Using facial characteristics to get the rough area of the eye and mouth, then use border extraction method for eye location and calculations of status and use Fisher linear classifier to get lip position and calculate mouth status. Conduct a comprehensive evaluation driver fatigue by making the definition of input and output of fuzzy controller and fuzzy inference rules. The thesis introduced the software design proposal of driver fatigue detection system. And the capability and stability of detecting system is tested. The main work of the thesis could be summarized as follows:Firstly, improve the traditional edge detection method which have more noise and higher wrong detect rate. Proposed to locate the boundary extraction method, experimental results show this method is fast and high accuracy.Secondly, use of Kalman filtering method to achieve the object of tracing and noise filtering, speeding up the detection rate further to improve the detection accuracy.Thirdly, comprehensive Evaluation of driver fatigue proposed the concept of the development control rules by using fuzzy controller on the comprehensive evaluation of fatigue, leaving a certain tolerance to the rate of detection.更多还原