Robust tracking is a challenging problem, due to intrinsic appearance variability of objects caused by in-plane or out-plane rotation and extrinsic factors change such as illumination, occlusion, background clutter and local blur. A tracker based on a single cue may be robust to certain distractions but vulnerable to some others. Therefore, it is appealing to fuse multiple cues into one tracker. In this paper, we propose an adaptive models combination framework for visual tracking. The color cue, texture cue and global representation of object are fused into one tracker by combination of three individual models. Then a simple yet effective adaptive weights strategy is proposed for evaluating weights of different models based on their performance. Experiments are performed on some changeling video sequences, both public and our own, show that our proposed framework achieve good performance.
展开▼
