In recent years, deep convolutional neural networks (CNNs) have achieved great success in visual tracking. To learn discriminative representations, most of existing methods utilize information of image region category, namely target or background, and/or of target motion among consecutive frames. Although these methods demonstrated to be effective, they ignore the importance of the ranking relationship among samples, which is able to distinguish one positive sample better than another positive one or not. This is especially crucial for visual tracking because there is only one best target candidate among all positive candidates, which tightly bounds the target. In this paper, we propose to take advantage of the ranking relationship among positive samples to learn more discriminative features so as to distinguish closely similar target candidates. In addition, we also propose to make use of the normalized spatial location information to distinguish spatially neighboring candidates. Extensive experiments on challenging image sequences demonstrate the effectiveness of the proposed algorithm against several state-of-the-art methods.

近年来，深度卷积神经网络（CNN）在视觉跟踪方面取得了巨大的成功。为了学习区分表示，大多数现有方法利用图像区域类别的信息，即目标或背景，和/或连续帧之间的目标运动。尽管这些方法被证明是有效的，但它们忽略了样本之间排名关系的重要性，这种关系能够更好地区分一个阳性样本，而不是另一个阳性样本。这对于视觉跟踪尤为重要，因为在所有积极候选者中只有一个最佳目标候选者，它紧紧限制了目标。在本文中，我们建议利用正样本之间的排名关系来学习更多的判别特征，以区分紧密相似的目标候选者。此外，我们还建议利用归一化的空间位置信息来区分空间相邻的候选对象。在具有挑战性的图像序列上进行的大量实验证明了所提出算法针对几种最新方法的有效性。