【作者】 刘爱军；

【导师】 谢公福； 梁德群；

【作者基本信息】 大连海事大学 ， 信号与信息处理， 2003， 硕士

【摘要】 运动目标跟踪是计算机视觉领域非常活跃的一个课题，而且有着非常广泛的应用，例如视频会议、机器人导航、虚拟现实等等。在会议电视系统中，大多数都装有可控制摄像机，对发言人的运动进行跟踪。现有的跟踪系统大部分是基于硬件实现的，使用时操作复杂而且价格昂贵，本文用软件的方法实现了对摄像机的控制，完成了会议电视系统中的图像跟踪子系统的开发。 另外，运动目标跟踪算法的好坏直接影响着运动目标跟踪的准确性和稳定性，现有的目标跟踪算法大多数都是基于特征或运动信息的，虽然能够完成对运动目标的可靠跟踪，但是需要处理的数据量大、运算复杂，很难达到实时跟踪的要求。本文在对目标检测和跟踪算法分析的基础上，详细研究了两种目标跟踪算法： (1)基于倒谱的目标跟踪算法，通过对目标图像对的倒谱进行分析，找出倒谱域中冲击函数的位置进而确定图像对中目标的位移大小，实现对运动目标的跟踪。另外用Hartley变换代替Fourier变换，实现了倒谱的快速算法。 (2)基于颜色直方图信息的目标跟踪算法，通过利用目标的特征颜色信息，得到当前帧颜色直方图的投影图，通过投影图的水平和垂直投影确定目标的位置，实现对运动目标的实时跟踪。更多还原

【Abstract】 Moving objects tracking is an active problem in computer vision and has a wide variety of applications, such as videoconference, robotics navigation, virtual reality etc. The controllable camera is used to track the speaker in most videoconference systems. Most of the tracking systems are based on hardware. However , hardware-based tracking system is sophisticated and expensive. In this paper, a software method is proposed for controlling the camera and the image-based tracking system is developed as a part of videoconference systems.In addition, the accuracy and stability of objects tracking depend on moving objects tracking algorithms to a great extent. Most of the current approaches proposed are mainly feature-based or motion-based. These approaches are reliable to track objects, but they have a lot of data to be processed and they are too complicated to apply in real time. In this paper, two tracking algorithms are researched in details by analyzing the current algorithms of objects detecting and tracking.(1) Cepstrum-based tracking algorithm: In this method, the impulse’ s position in cepstrum fields is computed and the object’ s position is accordingly obtained, thus, the object can be tracked in each frame. In addition, the usage of Hartley transform gives faster computational speed than Fourier transform.(2) Color histogram-based tracking algorithm: In this method, the projection of the color histogram is obtained by the main color of the objects. The object’ s position is computed according to the horizontal and vertical projections of projected image.更多还原