基于视频的多视角人脸检测与跟踪-模式识别与智能系统专业论文.docx

上海交通大学硕士学位论文 上海交通大学硕士学位论文 上海交通大学硕士学位论文 上海交通大学硕士学位论文 算法，实验结果表明本文的算法对于视频多视角人脸具有较高的 检测率，检测速度快，且鲁棒性高。 (3) 在目标跟踪的基础上，针对多视角人脸跟踪问题，本文研究了基于 粒子滤波的多视觉特征融合，多模型融合跟踪算法，并提出了一 种基于子空间特征模型，结合人脸离线模型并自动进行在线学习 的多视角人脸跟踪算法。在跟踪过程中，算法采用了一种新的自 适应粒子滤波跟踪框架，人脸模型使用子空间特征模型，并对人 脸的五个角度进行特征建模。实验结果表明，该算法能够跟踪多 视角变尺度人脸，并实时分辨人脸姿态，对人脸的旋转，尺度变 化，环境影响不敏感，具有较强的鲁棒性和精确性。 (4) 针对监控人脸识别系统的关键问题，本文研究并实现了最佳正面人 脸识别图像选择算法，包括小波域图像评价算法、基于子空间降 维的姿态估计算法、基于距离的人脸选择算法、基于 SVM 的人脸 分类算法。本文提出了一种新颖的 PTZ 摄像机获取感兴趣区域高 分辨率图像方法，设计实现了一套基于多任务决策融合编程架构 的监控人脸识别系统，进行了研究论证和技术演示。 关键字：多视角人脸检测，人脸图像规范化，多视角人脸跟踪， 子空间特征模型，多模型融合，自适应粒子滤波，在线学习，监控视 频人脸识别 VIDEO-BASED MULTI-VIEW FACE DETECTION AND TRACKING2 ABSTRACT Multi-view face detection and tracking is a significant aspect in computer vision and pattern recognition research. It is the base of face information processing and plays an important role in face recognition, man-machine interaction, video conference, 3-G mobile communication. This thesis mainly study on key technologies of video-based multi-view face detection and tracking, including video pre-process technique, multi-view face detection algorithm, face image standardization, multi-view face tracking and the realization of surveillance face recognition system. The main contribution of this dissertation is listed as follows. (1).Video preprocess algorithm including light compensation, image de-noise and super-resolution algorithm to satisfy the demand of real-time video and surveillance system. (2).Based on the existent face detection algorithm, the author realizes 2 This thesis is supported by a grant from the National High Technology Research and Development Program of China (863 Program)(No.2007AA01Z164）and National Natural Science Foundation of China(No. multi-view face detection and proposes face image validation and re-location algorithm using face features and color model. (3).This paper proposes an innovative algorithm based on sub-space model for robust multi-view face tracking under complex environment.