EACOFT: An energy-aware correlation filter for visual tracking

• Energy-aware-correlation-filter tracker to adaptively adjust the target for tracking. New strategy reject low quality samples and ensure model discriminant ability. Combining bottom-up top-down optimal training robust Outperform many state-of-the-art trackers on several challenging datasets. Correlation filter based attribute its calculation in frequency domain can efficiently locate targets a relatively fast speed. This characteristic however also limits generalization some specific scenarios. The reasons that they still fail achieve superior performance (SOTA) are possibly due two main aspects. first is while tracking objects whose energy lower than background, may occur drift or even lose target. second biased be inevitably selected training, which easily lead inaccurate To tackle these shortcomings, novel energy-aware correlation (EACOFT) method proposed, our approach between foreground background balanced, enables of interest always having higher background. samples’ qualities evaluated real time, ensures used template helpful with In addition, we propose an combined plays important role improving both effectiveness robustness As result, achieves great improvement basis baseline tracker, especially under clutter motion challenges. Extensive experiments over multiple benchmarks demonstrate proposed methodology comparison number SOTA trackers.