RAPID VISUAL TRACKING WITH MODIFIED ON-LINE BOOSTING AND TEMPLATE MATCHING

摘要

On-line learning is increasingly popular in visual tracking,rnbut the challenge that it faced is how to adapt thernappearance changes and avoid the drifting or missingrntrack. In this paper, a fast visual tracking algorithm isrnproposed to make the tracker more accurate and stable inrnthe complex variations situations like occlusions,rnilluminations and shape deformations. In the proposedrnalgorithm, a modified on-line boosting method isrndeveloped to make the tracker more adaptive to variablernscene and a template matching model is used to constrainrnthe training samples, so that the accumulating errors inrnself-update learning can be alleviated effectively. Inrnaddition, an optimization process is used to reduce therncomputational burden. Our experimental results haverndemonstrated that compared with other on-line trackingrnmethods the target can be accurately tracked with lowerrndrifting error in the complicated environments by usingrnthe proposed algorithm. Moreover, the new tracker runs atrn60 frames per second, and is suitable for the real-timerncatching and tracking.