TLD框架下的多人脸目标跟踪算法设计与实现-软件工程专业论文.docxVIP

三峡大学硕士学位论文 三 峡 大 学 硕 士 学 位 论 文 PAGE VI万方数据 PAGE VI 万方数据 适应多人脸跟踪领域的特殊需求。同时，结合实时人脸检测算法以及有效的多人脸的 标记与分离方案，最终实现多人脸的自动检测与跟踪。 5) 通过大量实验对本文所提出的多人脸跟踪算法的有效性进行了验证，同时对 实验结果进行了详细分析。 6) 最后利用相关编程语言和 OpenCV 计算机视觉库对该算法进行了系统实现。 关键词：计算机视觉 目标跟踪 人脸跟踪 智能监控 TLD Abstract With the development of computer and information technology, multi-face tracking has been widely studied in the practical application, such as intelligent video surveillance, human-computer interaction, intelligent traffic, virtual reality etc., and attracted a large number of domestic and foreign researchers. Such systems need to automate more complex environment for real-time monitoring and personal behavior analysis, the completion of this series of tasks need to involve a lot of technology in the field of computer vision, and people face tracking is one of the core technology. Multi-face tracking aims to simulate human motion sensing capabilities, enables a computer to automatically recognize the moving human face in the video sequences. Multi-face tracking is more complicated than generally tracking problem, because multi-face tracking not only to distinguish between foreground and background, but also distinguish different faces. Because of the similarity between human face led to great difficulty for the research. Although a large number of face tracking algorithm has been proposed now, but real-time and robust multi-face tracking is still a very challenging problem. This thesis focuses on the multi-face tracking problem. First, the state of the art in this field at home and abroad was discussed, and the difficulties of this project. This thesis uses a new and effective long time tracking framework (TLD) which combines traditional tracking, detection and learning to solve the target deformation, partial occlusions and other issues while tracking. Then this thesis based on TLD, and combined with real-time face detection algorithm proposes an automatic multi-face detection and tracking algorithm. The research of the