【作者】 王小明；

【导师】 黄昶；

【作者基本信息】 华东师范大学 ， 通信与信息系统， 2007， 硕士

【摘要】 人脸姿态估计和跟踪(Face Pose Estimation and Tracking)是指在摄像机获取的人脸图像序列中确定人脸在三维空间中姿态的方法。人脸姿态估计和跟踪作为智能人机交互和计算机视觉中的一个重要的研究课题，近年来正不断引起人们的兴趣。人脸姿态估计和跟踪在人脸识别、视频会议、智能监控、疲劳检测、虚拟现实、游戏和娱乐等方面都具有广阔的应用前景。人脸姿态估计和跟踪涉及到了人脸检测、人脸跟踪和人脸姿态估计等相关课题，目前研究人脸姿态估计和跟踪的方法包括基于人脸特征的方法和基于模型的方法，本文重点对基于模型的方法进行了研究。在分析比较现有算法的基础上，本文首先建立了混合肤色模型从背景中确定人脸位置，并提取人脸椭圆轮廓。然后利用改进的基于颜色信息的人脸特征点定位算法确定了双眼和嘴巴的位置。再根据视频序列的帧间相关性对人脸特征点进行实时跟踪。在获取人脸特征点的基础上，运用特征点之间的几何关系和人脸椭圆轮廓，结合人脸正面平行帧的信息；给出了估计人脸绕X、Y、Z轴旋转角度的算法，实现了对人脸三维姿态的估计和跟踪。最后，针对单目视频头肩像序列的情况，利用Delphi和OpenGL实现了一个实时的人脸姿态估计和跟踪系统，从而验证了本文算法的可行性，并提出了进一步的研究方向。更多还原

【Abstract】 Face pose estimation and tracking is the method to estimate 3D pose parameters from image sequence with human face. As a important research topic in the field of intelligent human computer interaction and computer vision, face pose estimation and tracking is becoming more and more attractive. The technology of face pose estimation and tracking can be widely applied in face recognition, video conference, intelligent surveillance, fatigue detection, virtual reality, game and entertainments.Face pose estimation and tracking relates to the field of face detection, face tracking, face pose estimation and so on, The existing approaches for face pose estimation and tracking can be classified into two categories: property-based and model-based, this dissertation emphasizes on the model-based method. After analysing the existing arithmetic, this dissertation first establishes hybrid skin model to locate face, and extract the face ellipse. then utilizes the arithmetic of improved face feature location based on the color information to determine the eyes and mouth position. According to the relation of video inter-frame, the face features is tracked, on the basis of acquiring face features, utilizing the face ellipse profile and geometric relationship among face features, combining the information of fronto-parallel face, the algorithm of face rotation angles around the X, Y and Z-axis is presented, so the 3D face pose estimation and tracking is implemented. Finally, this dissertation deals with the case with the monocular head-and-shoulder image sequence, using Delphi and OpenGL to implement a system of real-time face pose estimation and tracking for examining the effectiveness of our approach, and putting forward the directions for further research.更多还原