为了进一步提高目标跟踪算法中目标定位的精确度, 提出了一种基于多层卷积特征的目标跟踪算法。该算法首先利用VGG-Net-19的多层结构提取待测图像的多层卷积特征, 通过相关滤波方法获取多层卷积特征并对其进行加权融合, 从而确定目标的真实位置。然后通过结合多层卷积层以及全连接层的特征, 在目标表示效果上有明显提升, 在保证跟踪效率的同时提高精确度。实验结果表明, 与目前主流的HCF、MEEM、KCF、Struck四种目标跟踪算法对比, 该算法取得了优于其他方法的精度与成功率, 距离精确率提高了2～20%, 与OPE、SRE以及TRE的结果具有一致性。

In order to improve the accuracy and robustness of object tracking, a novel tracking algorithm based on hierarchical convolutional features is proposed. Firstly, using VGG-Net-19 networks to extract the hierarchical convolutional features of images has a greater improvement than using only one layer to do that. Secondly, the algorithm obtains features by using correlation filtering method with weight fusion , so as to determine the real position of the target according to the characteristics of different layers. By combining the feature of multiple convolutional layers and one full-connected layer, the algorithm can improve the target representation and tracking accuracy while ensuring the tracking efficiency. The experimental results show that compared with the current four popular tracking algorithms, including HCF, MEEM, KCF and Struck, the algorithm achieves better accuracy and success rate than the other methods, and distance accuracy is improved 2-20%. At the same time, the tracking results are consistent in OPE, SRR and TRE.