Most of existing correlation filter-based tracking approaches only estimatesimple axis-aligned bounding boxes, and very few of them is capable ofrecovering the underlying similarity transformation. To tackle this challengingproblem, in this paper, we propose a new correlation filter-based tracker witha novel robust estimation of similarity transformation on the largedisplacements. In order to efficiently search in such a large 4-DoF space inreal-time, we formulate the problem into two 2-DoF sub-problems and apply anefficient Block Coordinates Descent solver to optimize the estimation result.Specifically, we employ an efficient phase correlation scheme to deal with bothscale and rotation changes simultaneously in log-polar coordinates. Moreover, avariant of correlation filter is used to predict the translational motionindividually. Our experimental results demonstrate that the proposed trackerachieves very promising prediction performance compared with thestate-of-the-art visual object tracking methods while still retaining theadvantages of high efficiency and simplicity in conventional correlationfilter-based tracking methods.
translated by 谷歌翻译