导　　师： 秦华标

学科专业： 080902

授予学位： 硕士

作　　者： ;

机构地区： 华南理工大学

摘　　要： 随着汽车消费的普及汽车消费走进了千家万户,交通安全问题也受到了社会普遍关注成为热点。疲劳驾驶是当前导致车祸事故主要的原因之一。驾驶员疲劳检测系统的研究已受到世界各大汽车厂家的关注,具有重大的社会意义和良好的市场前景,而该类产品的原理和技术不成熟性,则给我们提供了很大的研究和发展空间。 在各种疲劳检测方法中,基于PERCLOS/(percentage of eyelid closure/)的疲劳检测方法是最实用和可靠的,该方法的关键点就是驾驶环境中实时、准确地人眼睁闭状态跟踪。为此本文提出了一种新颖的人眼状态判别算法,该方法比传统方法简单,且对左右眼都做睁眼图像检测,因而对睁眼图像的人眼状态判别准确率更高。 本文在对Viola人脸检测的特征分类器架构详细分析研究的基础上,提出了一种新的特征分类器训练方法,使用该方法对搜集的睁眼图像反刍样本集进行增量迭代训练后,得到了性能优秀的睁眼图像检测分类器,其睁眼图像检测准确率高达94.62/%。本文的驾驶环境中人眼睁闭状态跟踪算法在实时性和准确性两个方面都有着良好的性能,对960张320*240包含睁闭眼的彩色图像完成判别平均耗时18.8毫秒,判别准确率高达98.12/%,算法处理帧速高达53帧//秒。 最后本文完成了跨操作系统平台人眼睁闭状态跟踪算法的代码实现,并将该算法在列车驾驶员辅助系统中应用,该系统在嵌入式车载硬件平台运行时,取得了平均18帧//秒的运行速度,且能快速准确地统计出PERCLOS值,对疲劳状况做出及时的报警。 Fatigue driving is currently one of the leading causes of car accidents. The driver’s fatigue detection system is designed for the real-time monitoring of drivers, to avoid accidents it will alarm the driver, when he is drowsing. Study of these products has been the concern of the world's major automotive manufacturers. The fatigue detecting method based on PERCLOS /(percentage of eyelid closure/) is the most practical and reliable in a variety of methods of fatigue detection. The key of this method is real-time eye state tracking under driving conditions. Eye state detection is a complex classification problem. This paper presents a novel algorithm of eye state detection. This method is simpler than traditional method; first it detects face from the input image, gets the results of eye state by directly detecting open eyes on face images. Face detection and open eye detection is two main algorithm of this article. In essence, they are target object detection on the input images. This article proposes a new feature classifier training method based on detailed analysis of the feature classifier architecture of Viola face detection. In the research, after the incremental iterative training of collected ruminant sample sets of open eye images, we’ve got an excellent classifier of open eye image detection classifier. Open eye image detection accuracy of the classifier reaches 94.62/%. The speed and accuracy of eye state detection algorithm are two difficulty problems, Previous algorithm is difficult to have both speed and accuracy. The proposed algorithm makes full use of face image information, and completes eye state identification simply and accurately by the classifier of open eyes image detection. From the tested results in this paper, completing eye state identification on 960 color images/(320 * 240Resolution/) took an average processing time of 18.8 ms, with identification accuracy rate of 98.12/% and frame speed up to 53 frames per second. This shows that the proposed eye state detection algorithm can achieves high accuracy and speed. Finally, source code implementation of eye state tracking algorithm is completed, which can be compiled cross-operating system.

关 键 词： 驾驶员疲劳检测 人眼状态检测 人脸检测

分 类 号： [TP391.41]

领　　域： [自动化与计算机技术] [自动化与计算机技术]