基于视觉的眼动特征研究

发布时间：2018-03-25 23:17

本文选题：眼动特征　切入点：眼睑匹配　出处：《北京交通大学》2017年硕士论文

【摘要】：眼睛是我们最重要的特征之一,我们通过眼睛来获取外界信息。眼动特征研究在疲劳驾驶、认识视觉工作的原理、分析人的情感及行为、基于眼动的人机交互等问题上都发挥着关键作用。基于视觉的眼动特征技术具有易操作性、低损失、高准确度等优势,目前是眼动研究的主要方向。在过去的几十年中,眼动特征研究虽然取得了显著的进步,但仍有很多不可控制的因素,影响检测结果。每个人的眼睛形状和大小都不同,眼睑遮挡眼睛的面积也不一样,位置和光照条件变化的不同,睫毛等噪声的干扰,都会直接影响到眼动特征研究的准确性,市面上单纯采集眼部数据设备价格非常昂贵且数量稀少,大部分的眼动特征研究都是建立在人脸检测的基础上,直接对眼睛进行检测和特征分析十分稀缺。鉴于这些问题,本文集中研究了眼睑和瞳孔两部分眼动特征。首先提出了一个基于ASM算法和Kalman滤波的人眼检测模型——AK-EYE模型,并应用该模型对实时眼睛形状和位置跟踪算法进行改进,提高眼睑匹配的速度和精度;然后,通过模板匹配技术来定位瞳孔,并专门采集眼部数据集进行验证实验;最后,结合眼动特征对疲劳检测方法进行实验分析。主要内容如下:(1)详细分析了相关的眼动特征研究方法的主要思想及其优缺点,在眼睑匹配和定位过程中,针对ASM算法中出现的匹配位置不准确的问题,提出了结合ASM算法和Kalman滤波的AK-EYE模型,并利用该模型对眼睑进行匹配。首先,按照特定标准选取合适的样本特征进行标定,建立眼睑形状模型,然后,对眼睑轮廓进行搜索匹配,预测和更新模型的初始位置,完成实时的眼睑定位跟踪,并分析实验结果,验证算法的有效性。(2)针对包含人脸的数据集和单纯眼部的数据集,选择合适的模板,利用模板匹配算法实现瞳孔的精确定位,并分别进行实验结果展示和分析。对不同的模板匹配算法性能进行比较,并将本文模板匹配算法在公共数据库上进行实验验证。(3)根据眼动特征研究得到的眼睑信息和瞳孔信息,建立疲劳预警系统。该疲劳预警系统使用眼睑特征信息和瞳孔特征信息作为疲劳状态的判断输入参数,并将这些眼动特征和PERCLOS测定原理结合,构建新的疲劳分析判断算法,判断疲劳状态。
[Abstract]:The eye is one of our most important features. We use our eyes to obtain information about the outside world. Eye movement features research fatigue driving, understanding the principles of visual work, and analyzing people's emotions and behaviors. Eye movement feature technology, which has the advantages of easy operation, low loss, high accuracy and so on, is the main research direction of eye movement research in the past few decades. Although significant progress has been made in the study of eye movement characteristics, there are still many uncontrollable factors that affect the test results. Each person's eyes are different in shape and size, and the area of eyelid occlusion is different. The difference of position and illumination condition, the interference of eyelash and other noise will directly affect the accuracy of the study of eye movement characteristics. The price of the simple collection of eye data is very expensive and the quantity is scarce. Most of the research on eye movement feature is based on face detection. It is very rare to directly detect and analyze the eye features. In view of these problems, This paper focuses on the eye movement characteristics of eyelid and pupil. Firstly, an eye detection model AK-EYE based on ASM algorithm and Kalman filter is proposed, and the real-time eye shape and position tracking algorithm is improved by this model. Improve the speed and accuracy of eyelid matching; then, through template matching technology to locate the pupil, and special collection of eye data set for verification experiment; finally, The main contents are as follows: (1) the main ideas and advantages and disadvantages of the related methods are analyzed in detail. In the process of eyelid matching and locating, In order to solve the problem of inaccurate matching position in ASM algorithm, a AK-EYE model combining ASM algorithm and Kalman filter is proposed, and the model is used to match eyelids. Firstly, appropriate sample features are selected to calibrate according to specific criteria. The eyelid shape model is established, then the eyelid contour is searched and matched, the initial position of the model is predicted and updated, the real time eyelid location tracking is completed, and the experimental results are analyzed. To verify the validity of the algorithm, we select the appropriate template for the dataset containing the face and the simple eye data set, and use the template matching algorithm to locate the pupil accurately. The performance of different template matching algorithms is compared, and the template matching algorithm of this paper is tested on the common database to verify the eyelid information and pupil information obtained from the study of eye movement characteristics. A fatigue early warning system is established, which uses eyelid characteristic information and pupil characteristic information as input parameters to judge fatigue state, and combines these eye movement characteristics with the principle of PERCLOS measurement to construct a new fatigue analysis and judgment algorithm. Judge the fatigue state.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】