Research on visual object tracking by dynamic and static metric
Lichao Zhang, BI Du-yan, Yufei Zha, Nan Dong, Li Zhou · 2014
Color feature is used in object model of classical kernel tracking algorithm, but the static feature is hard to adapt to the changing object with abrupt movement and rotation in the background whose color distribution is similar to object's. Considering that local dynamic feature extracted by classical optical flow can describe the object's dynamic characteristics, a new tracking method is proposed based on dynamic metric. In order to obtain accurate denotation, the variance matrix is used to estimate the error ellipse which is adaptive to the object's scale and rotation. The experiments show that the proposed algorithm can achieve better tracking results compared with other related algorithms when the target moves abruptly and rotates under the condition of complicated background.