基于在线学习的目标跟踪方法研究*

针对视频目标跟踪问题，提出了一种基于co-training框架下的在线学习跟踪方法。该方法首先根据两种不同的局部特征，利用在线 Boosting算法分别建立模型, 然后采用co-training框架来协同训练，有效避免了模型误差累积和跟踪丢帧等问题。实验证明了该方法的有效性。

Object tracking research based on on-line learning

To video object tracking problem, this paper proposed an on-line learning tracking method based on co-training framework. First of all, the method adopted two different local features to build on-line Boosting model, and then, would train samples making use of co-training learning framework, which avoided the cumulative error of the model and dropping frames problem effectively. Furthermore, some experiments have been maded and the results implyed that the new method is very efficient.