利用区域变形和背景更新实现运动对象跟踪

从时域统计的角度出发，提出了一种结合自适应混合背景更新模型的区域变形跟踪算法．该算法以模型更新得到的前景／背景二值分割掩膜作为区域特征，将跟踪问题抽象为一个水平集(Level Set)偏微分方程的数值求解问题，并分析了算法的自适应性．为了进一步提高算法的实现效率，引入了窄带跟踪方案．实验表明，该算法可以对视频序列中的指定运动对象进行快速精确的跟踪。

Moving Object Tracking Based on Region Deformation and Background Modeling

We address the problem of tracking moving objects through a video sequence using deformable regions. So far, numerous tracking approaches based on region deformation have been proposed and the chosen region-based features are mostly special visual information such as mean-color and texture. This kind of information may be misleading in some dense visual clutter. In this paper, we present a temporal statistical tracking algorithm that extracts the region-based features via adaptive background mixture model. Our contribution is to define a new tracking criterion combining geometrical and motional features of the region. The resultant algorithm is expressed in the form of a level set partial differential equation. We also analyzed the adaptive ness of our algorithm. To further improve the computational efficiency, we introduce the narrowband method into our tracking scheme. Finally, highly promising experimental results are provided to illustrate the efficiency and accuracy of our method.