Research on visual object tracking by fusing dynamic and static features

Traditionally, most tracking algorithms only use the single static feature or single dynamic feature to model the object. The static feature based model can not describe the object's dynamic characteristics and is difficult to adapt to the changing object with a background cluster, abrupt movement and rotations. While the classical optical flow is able to describe local dynamic characteristics, it has aperture issues. Therefore, we present a new tracking method based on fusing the dynamic and static features adaptively: the dynamic feature is extracted by the bidirectional optical flow and error metric adaptively, and is fused with the static feature by the fusion weight efficiently. The fusion weight based covariance is constructed to evaluate error ellipse which describes the object's scale and orientation exactly; the weight assignment parameter is updated by an on-line parameter updating mechanism, which balances the dynamic feature and static feature and ensures the tracking adaptation to the object's velocity and scene changes. Experiments show that the proposed algorithm can achieve better tracking results compared with the related algorithms, on the occasions when the object moves abruptly and rotates with a background cluster.