序列图像中人眼的定位与跟踪毕业论文.doc

本科毕业论文 (科研训练、毕业设计) 题 目：序列图像中人眼的定位与跟踪 姓 名： 学 院：软件学院 系：软件工程 专 业：软件工程 年 级： 学 号： 指导教师（校内）： 职称： 年月 [摘要] 当前的眼睛定位算法大多速度较慢，实时性不强。大多数快速的眼睛跟踪算法容易受到光照变化和姿态变化等的影响。针对这些问题，本文提出了一种在复杂背景，姿态变化条件下的实时人眼定位与跟踪算法。首先使用AdaBoost算法训练眼睛检测分类器，用这个层叠检测分类器粗略检测人眼区域，得到多个眼睛点；然后利用人眼的几何特征建立规则确定人眼对的位置与大小并找出眼球的位置；接着以眼球为初始窗口，应用CamShift算法对眼睛进行跟踪。当姿势变化或光照变化引起CamShift跟踪失效时，再次应用分类器检测确认眼睛点，继续跟踪。本算法的主要特点是利用几何特征创建规则确定眼睛位置以及跟踪丢失时的隔帧重新检测眼睛的算法。我采用VC6.0开发环境，使用MFC和OpenCV类库，开发了一个序列图像人眼跟踪实验程序。实验证明，本算法可以有效地，快速地跟踪人眼，并且有一定的鲁棒性，可以应用于眼睛信息提取，视线跟踪，疲劳检测等许多领域。 AdaBoost CamShift Human Eyes Detection and Tracking of Image Sequences [Abstract] Currently most eyes detection algorithms are so slow that cannot be applied to real-time environment, while most rapid eyes tracking algorithms are easily impacted by brightness variation and orientation variation. In order to solve such limitations, a real-time human eyes detection and tracking method under complex background and face orientation variation is proposed. First the eyes detector is trained by AdaBoost algorithm, and then this cascade detection classifier is used to detect the candidate eye pair areas. Then a rule is established according to the geometric features of human eyes to validate the exact eyes pair’s position and size and find out the position of the eyeballs. Finally use the eyeball as the initial search window, apply the CamShift to the window to track the eyes. When the tracking is missed because of orientation change of illumination change, the detector is applied again to find out the eyes, and then continue tracking. The main innovation of this method is validating the eyes position by geometric rules and the algorithm to redetect eyes after a few frames. Human eyes tracking program is developed under VC6.0 integrated development environment, using MFC and OpenCV library. Experiments show that the method can effectively and rapidly track the human eyes while it’s robotic. This method can be used in eyes information extraction, eye