Long-term tracking is one of the most challenging problems in computer vision. In this paper, we make full use of the Discriminative Correlation Filter (DCF), and propose a real-time long-term tracker by exploiting a joint tracking–verification–detection–refinement framework. We utilize a DCF which is updated aggressively to estimate translation and scale variation of the target. Subsequently, a passively updated DCF checks the reliability of the tracking result. Once the result is not reliable, we evoke the proposed optimized candidate detector to generate a small number of relatively high quality candidates. Finally, one DCF with an adaptive online learning rate is adopted to refine the predictions that the sparse candidates inferred. In addition, we employ a selection mechanism for the correlation responses to maintain reliable samples effectively. Extensive experiments show that the proposed method performs favorably against lots of state-of-the-art methods while running more than 30 frames per second on single CPU.

长期跟踪是计算机视觉中最具挑战性的问题之一。在本文中，我们充分利用了判别相关滤波器（DCF），并通过利用联合跟踪，验证，检测和优化框架，提出了一种实时长期跟踪器。我们利用DCF，该DCF会积极更新以估计目标的平移和比例变化。随后，被动更新的DCF检查跟踪结果的可靠性。一旦结果不可靠，我们就会提出建议的优化候选检测器，以生成少量相对高质量的候选。最后，采用具有自适应在线学习率的DCF来改进稀疏候选者推断的预测。此外，我们对相关响应采用选择机制，以有效地维护可靠的样本。大量实验表明，所提出的方法在单个CPU上以每秒30帧以上的速度运行时，相对于许多最新方法具有良好的性能。