The technology of visual tracking has promising future in applications of video surveillance, intelligent traffic control, human-computer interaction, video compress etc. It is a popular research topic in Computer Vision. In this dissertation, we focus on the fundamental research of classification based tracking algorithm. Through reviewing recent tracking algorithms, we propose two classification based algorithms. And the content of this dissertation is summarized as below. Firstly, most of tracking algorithms, especially those published in recent five years are reviewed. Some are analyzed with both advantages and disadvantages, which helps us to make improvements and devise our own algorithms. Secondly, under the classification based tracking framework, we focus on constructing accurate and robust classifiers. Two methods are proposed, one is boosting the relative spaces, and the other is based on local distance learning. Thirdly, a group of criteria for measuring the accuracy and stability of tracking algorithm is proposed initially. We do experiments on many sequences with different tracking algorithms. Through comparison based on our criteria, we show the superior performance of our tracking algorithms.
修改评论