Abstract: The Correlation Filter（CF） tracking algorithms have achieved outstanding performance by using efficient discriminative regression model and multi-cue features, such as Histograms of Oriented Gradients（HOG） and Color Names（CN）. However, the performance still suffers from insufficient discriminative information during appearance variations. To mitigate this problem, a Self-Learning based Discriminative Correlation Filter tracking algorithm（SLDCF） is proposed. The self-learning feature is obtained by exploring the collaborative representations between successive frames. It extracts the information from target variation and alleviates the impact from background. The experimental results on the standard video benchmarking dataset demonstrate the effectiveness and robustness of the proposed algorithm and its superior performance in comparison with other traditional correlation filter tracking algorithms.

Key words: discriminative regression model, multi-cue features, histograms of oriented gradients, color names, correlation filter tracking algorithms, self-learning features

摘要： 依靠高效的鉴别回归模型和多线索特征，如方向梯度直方图（HOG）特征和颜色名（CN）特征，相关滤波（CF）跟踪算法取得了优异的跟踪效果。但其弱点是不能应对由表观变化过程中鉴别信息不充分而导致的跟踪失败。针对这一问题，提出了基于自学习特征的相关滤波跟踪算法（SLDCF）。其中，自学习特征探索了相邻帧之间协同表示的特性，能够学习到相邻帧之间的目标变化情况，同时有效减少背景的干扰，以提高滤波器的鉴别性。通过标准视频数据集上的验证对比实验，其跟踪效果优于其余传统的相关滤波跟踪算法，证明了该算法的有效性和鲁棒性。

关键词: 鉴别回归模型, 多线索特征, 方向梯度直方图, 颜色名, 相关滤波跟踪算法, 自学习特征