Bounding-box Centralization for Improving SiamFC++

Object tracking is one of the important tasks in the field of computer vision. SiamFC++is an excellent anchor-free single object tracking method. It has high accuracy but low success rate. To improve the success rate of the tracker and enhance its robustness, this paper proposes a bounding-box centralization strategy (BCS) to deal with the problem of low success rate. In this paper, two new bounding boxes of elliptical bounding-box (EB) and rectangular bounding-box (RB) are designed to improve SiamFC++, and tested on OTB and VOT datasets. The test results demonstrate that BCS can improve the performance of SiamFC++.