基于核相关滤波器的目标跟踪算法

Visual tracking algorithm based on kernelized correlation filters

针对目标跟踪过程中出现的目标形变、遮挡、出平面旋转等干扰问题,通过对传统核滤波相关(KCF)跟踪算法在特征提取方式和模型更新方案上的改进,提出一种基于颜色名称空间特征的核相关滤波算法。为了验证算法的有效性,在目标跟踪标准数据集中选取了38个彩色视频序列对跟踪算法进行实验验证,并同时与KCF、Struck、TLD、SCM等优秀目标跟踪算法进行对比。实验结果表明所提出的新算法不仅具有最好的跟踪效果,同时在目标形变、遮挡、出平面旋转等干扰条件下具有更好的适应性。

To address the disturbance caused by object deformation, occlusion and out- of-plane rotation in visual tracking, this paper proposed a new kernelized correlation filtering algorithm on the basis of color name and space features by improving the traditional kernelized correlation filters（KCF） tracking algorithm on the aspects of feature extraction mode and model updating scheme. In order to verify the effectiveness of the algorithm, 38 color video se-quences were selected in visual benchmark datasets for verifying the tracking algorithm. In addition, the paper com-pared the performance of the algorithm with other competitive visual tracking algorithms such as KCF, structured output tracking with kernel （ Struck） , tracking- learning-detection （TLD） and sparsity-based collaborative model （SCM）. Results show the proposed algorithm not only has the best performance, but also is more adaptive to track-ing challenges such as deformation,occlusion and out- of-plane rotation.