【作者】 沈文忠；

【导师】 何小卫； 梁久祯；

【作者基本信息】 浙江师范大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 目前交通事故中相当一部分是由司机疲劳驾驶引起的,因此,对驾驶员的疲劳状态进行检测,减少此类事故的发生,有着积极的现实意义。本文对疲劳检测的算法进行了一定研究,通过研究司机的眼睑运动即眼睛闭合状态和凝视方向来综合评判司机的疲劳程度。本文主要做了以下几个方面的工作:(1)为了提高人眼定位的准确度和速度,本文采用ASM(主动形状模型)进行人眼的粗定位,然后单独对人眼区域进行肤色分割,实现人眼细定位,细定位的结果是得到5个人眼特征点。实验表明该方法速度快,准确率高。(2)为了达到实时性要求,定位之后就要进行人眼的跟踪,本文提出了一种新颖的眼睛跟踪方法,即综合Kalman滤波和ASM算法来进行人眼跟踪。整个跟踪过程分为两个阶段:首先根据上一帧图像中人眼的位置运用Kalman滤波预测在下一帧中人眼的位置,然后以此作为ASM的初始状态,跟踪人眼。(3)根据跟踪到的人眼进行眼睛闭合状态识别和凝视方向计算。本文提出新颖的方法即对人眼区域的肤色分割图计算人眼黑色块面积的方法来计算人眼闭合程度;对于凝视方向,本文通过获取的5个人眼特征点以几何的方法进行计算,将凝视方向分为左中右,上中下共9个方向以进行区分,方法简单、快速、效果好。(4)根据单一参数评判疲劳程度效果并不是很好,本文综合人眼闭合程度和凝视方向进行司机驾驶疲劳的检测,建立新的评判规则。本文的创新点主要有以下三个方面:(1)虽然以前也有人在疲劳检测的时候提到人眼凝视方向,但并没有真正指明如何计算人眼凝视方向,并提出评判规则。本文用图像处理的方法计算人眼凝视方向,并与PERCLOS一起建立疲劳检测的评判规则。(2)在人眼定位与跟踪算法上,本文将ASM与基于肤色的方法相结合,速度快、效果好。(3)相对于通过模板匹配的方法计算人眼闭合程度,本文所采用的通过计算人眼黑色块的面积的方法更加简洁明了,可靠性更高。更多还原

【Abstract】 Nowadays, many traffic accidents take place because of the fatigued drivers. Therefore, it is significant to detect drivers’ fatigue state to decrease these traffic accidents. In this thesis, fatigue detection methods have been studied and driver’s fatigue level has been estimated according to his eyelid movement (eye state) and gaze direction synthetically.In this thesis four facets work has been done.Firstly, in order to improve the veracity and speed of eye detection, ASM has been used for coarse eye detection and then segments the eye region alone for careful eye detection, the purpose of careful eye detection is to obtain five eye feature points. The Experiments indicate that this method is fast and accurate.Secondly, for the purpose of real time ability, this thesis proposes a novel eyes tracking method after the detecting of the eyes, which synthesizes Kalman and ASM arithmetic. The whole tracking process is separated into two steps: first of all, using Kalman to estimate the location of the eyes in the next frame according to the location in the previous frame, and then take it as the initial state of the ASM to track the eyes.Thirdly, calculates eye state and gaze direction according to the results of eye detection. This thesis proposes a novel method to get the eye state by calculating the eye black block area in segmentation image. Gaze direction has been divided into nine parts, each part can be calculated by geometrical method according to five eye feature points, which has several advantages such as briefness、celerity、better effect.Fourthly, using single parameter to estimate fatigue level doesn’t work well, so this thesis combines eye status and gaze direction to create new formulas for driver fatigue detection.This thesis has three innovations:Though gaze direction was proposed before in fatigue detection, the way to calculate gaze direction didn’t designate clearly. The method of image processing has been used to calculate the gaze direction and then createfatigue detection formulas with PERCLOS.This thesis proposes a novel method to combine ASM and skin color segmentation for eye detecting and tracking, which is faster and better than others.Compared with the method of using template matching to calculate eye status, the method this thesis used is more succinct and credible, which calculates the area of eye black block.更多还原