摘要：本文主要实现了一种鲁棒性强、执行效率高的基于相关滤波器的长期跟踪算法，它 的主要特点在于将跟踪任务分解为对目标的位移估计和尺度变化估计，同时与补充性重 检测方案相结合。本文首先介绍基于相关滤波器的主流跟踪算法，并就一些跟踪器的原 理做出详细阐述。接着主要论述本文要实现的基于相关滤波器的长期跟踪算法原理和框 架，包括从一个图像帧中训练两个不同的基于相关滤波器的回归模型，其中通过使用时 间上下文的相关性来估计平移，通过判别式相关滤波器来进行尺度估计。并且将实现的 算法在标准数据集中运行测试，与当前最先进的跟踪器进行比较。除此之外，还实现了 一个以该算法为核心，可以用于实时视频跟踪的交互式应用。最后，分析了长期跟踪算 法中还需要解决的问题和未来的发展趋势。

关键词 目标跟踪；长期跟踪；相关滤波器；

Title Research on Long-term Tracking Based on Correlation Filter

Abstract：This paper is mainly about the implementation of a long-term tracking algorithm based on correlation filters, which has strong robustness and high execution efficiency. It’s tracking task is decomposed into the estimation both of translation and scale change. Also, it has a supplementary re-test module. This paper first introduced the most popular tracking algorithms based on correlation filters, and elaborates the principle of some trackers. Then ,the paper focuses on the principles and framework of the long-term tracking algorithm based on correlation filters, including training two different regression models based on correlation filters from one image frame, in which the translation is estimated by using the temporal context correlation, the scale estimation is estimated by discriminant correlation filters. After this, this algorithm is tested by running on a standard data

set, with the current most advanced trackers to compare. In addition, it also implements an interactive application and can be used for real-time video tracking. Finally, this paper talks about some problems of long-term correlation trackers that need to be solved and its future trends.

Keywords target-tracking; long-term tracking; correlation filter;

目 次

1 绪论 3

1.1 背景及意义 3

1.2 相关技术的现状 4

1.3 总体技术方案及其社会影响 5

1.4 技术方案的经济因素分析 6

1.5 论文章节安排 6

2 视觉跟踪基础 7

2.1 视觉跟踪应用领域 7

2.2 视觉跟踪的基本原理与方法 7

2.3 相关滤波器介绍 10

3 基于相关滤波器的长期跟踪算法设计 12

3.1 基于相关滤波器的长期跟踪算法基本原理和特色 12

3.2 算法公式推导与理论说明