Improve Visual Tracking by End-to-end Multi-Tracker Selection

While many state-of-the-art trackers focus on solving certain parts of the tracking problem (e.g., dealing with occlusion and scale variation), it is beneficial to exploit the merits of multiple trackers to tackle different application scenarios. This paper proposes an end-to-end tracker selection method to improve visual tracking performance. Any trackers with arbitrary tracking strategy can be integrated in our framework without the need of modifying any single component tracker, which is simple and effective. We introduce a novel verification mechanism to select the most reliable tracking result from multiple trackers. Both qualitative and quantitative evaluations demonstrate that the proposed method achieves a consistent performance improvement on the benchmark dataset.