基于多阶段学习的相关滤波目标跟踪

发布时间：2018-03-21 20:53

本文选题：目标跟踪　切入点：多阶段学习　出处：《电子学报》2017年10期 　论文类型：期刊论文

【摘要】：由于跟踪过程中目标外观变化和遮挡因素的影响,采用单一迭代更新滤波器的KCF算法在学习过程中会积累过多的噪声信息导致目标丢失.为解决该问题,本文提出一种基于多阶段学习的相关滤波跟踪算法.通过建立具有互补关系的全局阶段滤波器模型、一致性阶段滤波器模型以及初始阶段滤波器模型并行的对目标进行跟踪.在benchmark数据集的51个视频上的实验表明,本文算法取得的总体精度得分77.6%和总体成功率得分68.9%优于现有的大部分跟踪算法.
[Abstract]:Because of the change of target appearance and the influence of occlusion factors in the tracking process, the KCF algorithm with a single iterative update filter will accumulate too much noise information during the learning process, which leads to the loss of the target. In this paper, a correlation filter tracking algorithm based on multi-stage learning is proposed. The consistent stage filter model and the initial stage filter model are used to track the target in parallel. The experiments on 51 videos of benchmark dataset show that, The total accuracy score and the total success rate score of this algorithm are 77.6% and 68.9% respectively, which are superior to most of the existing tracking algorithms.
【作者单位】：武汉大学计算机学院;
【基金】：国家自然科学基金(No.61471274)
【分类号】：TN713;TP391.41

【相似文献】