基于双CCD的人眼视线追踪系统研究与设计

发布时间：2018-03-30 10:17

本文选题：摄像机标定　切入点：立体视觉距　出处：《长春工业大学》2017年硕士论文

【摘要】：人机交互技术,主要含有所视部分的跟踪,声音的识别,手势的识别等等的各种技术融合在一起的一种技术。其中,所视部分的跟踪是这个领域中最新出现的技术引起了国内外研究者的浓厚兴趣,这项新技术的出现,吸引了人们的目光,也是此领域向前发展的项目之一。当前,所视部分的追踪设备多是基于红外光源来定位视线位置或者让被观察者戴着特别的头盔来检测眼部运动,对被观察者都有干扰。本文围绕着视线跟踪这个主题,提出了基于非红外光学摄像机的视线追踪技术。主要探讨了立体视觉标定、测距和人脸检测、特征点检测、视线方向的计算等等与此有关的话题。第一,简单的概括了一下人机交互技术的发展历程,和计算机视觉的相关理论形成的过程。其次,还简单的概括了一下所视部分的追踪技术的概念,和怎样运用这项技术,分析他的研究的成果,和未来的发展走向。与此同时,告诉大家本文为何研究此内容和怎样研究此内容。第二,探索了双目CCD立体视觉测标定及测距理论和摄像机的数学模型,总结了常用的摄像机标定方法,并对双目CCD系统测距建立数学模型,分析其系统误差。在此基础上,应用基于BP的双目CCD标定算法,作为测距系统的数学模型。再次,研究了人脸及人脸特征检测技术。首先利用人脸肤色模型,把人脸与背景区分开来,利用人脸的几何特征,排除与肤色相近的非人脸区域,得到人脸区域,在人脸区域内进行眼睛的粗定位。然后将图像灰度化,用Sobel算子求出灰度图像的梯度信息,再利用基于圆检测的梯度Hough变换快速定位出人眼,内眼角和嘴角的提取则采用SUSAN算子。最后,提出了视线方向计算方法。借助眼球的数学模型以及双目CCD测距技术,根据两内眼角和两嘴角建立人脸平面,并根据历史实验数据计算出眼球中心在人脸平面坐标系中的坐标,再计算出被观察着照片中人眼中心的坐标,那么眼球中心与人眼中心连线的方向就是所要求的视线方向。实验结果表明,平均的视线偏差角度为7.65°。
[Abstract]:Human-Computer interaction (HCI) is a kind of technology which includes tracking of the viewing part, recognition of sound, recognition of gesture and so on.Among them, the tracking of the viewing part is the new technology in this field, which has attracted the attention of people, and is also one of the forward development projects in this field.At present, most of the tracking devices are based on infrared light source to locate the position of the line of sight or to allow the observer to wear a special helmet to detect eye movement, which interferes with the observed.Focusing on the subject of line of sight tracking, this paper proposes a line of sight tracking technique based on non-infrared optical camera.This paper mainly discusses stereo vision calibration, ranging and face detection, feature point detection, line of sight calculation and so on.First, briefly summarizes the development of human-computer interaction technology, and the formation of computer vision theory.Secondly, it briefly summarizes the concept of tracking technology in the visual part, and how to use this technology to analyze the results of his research and the development trend in the future.At the same time, tell you why this article and how to study this content.Secondly, the theory of binocular CCD stereo vision calibration and ranging and the mathematical model of camera are explored, and the commonly used camera calibration methods are summarized. The mathematical model of binocular CCD system is established and its system error is analyzed.On this basis, the binocular CCD calibration algorithm based on BP is used as the mathematical model of the ranging system.Thirdly, face and face feature detection techniques are studied.Firstly, the facial color model is used to distinguish the face from the background, and the geometric features of the face are used to eliminate the non-face region similar to the skin color, and then the face region is obtained, and the coarse eye location is carried out in the face region.Then the grayscale image is grayscale and the gradient information of grayscale image is obtained by Sobel operator, then the human eye is located quickly by gradient Hough transform based on circle detection, and the SUSAN operator is used to extract the angle of inner eye and mouth.Finally, the calculation method of line of sight direction is put forward.With the help of the mathematical model of eyeball and binocular CCD ranging technology, the face plane is established according to the angle of two inner eyes and two corners of mouth, and the coordinates of eye center in face plane coordinate system are calculated according to the historical experimental data.Then calculate the coordinates of the center of the eye and the center of the eye, and the direction of the eye center is the direction of the line of sight.The experimental results show that the average angle of line of sight deviation is 7.65 掳.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】