摘要可视目标跟踪是计算机视觉领域中的一个热门研究方向，在交通、导航、军 事等领域具有广泛的应用前景。在 2009 年，Kalal 等人提出了一种高效的跟踪算 法—TLD (Tracking- Learning- Detection)算法。TLD 算法将跟踪、检测和学习结合 起来，具有很好的鲁棒性。TLD 使用了 LK 跟踪器，利用跟踪成功的点的中值来 跟踪目标，但是在目标变形、遮挡等情况下这些点并不能很好地表示目标。针对 这个问题，本文对 LK 跟踪器跟踪成功的点进行聚类，挑选出能较好代表目标的 点来进行跟踪，并在此基础上实现了基于 TLD 的多目标跟踪算法。改进后的算法 即保持了 TLD 的高效性，也在一定程度上提高了算法的准确度。最后对改进前后 的跟踪效果进行了对比分析，并提出了不足之处和改进方法。68464

毕业论文关键词 跟踪-学习-检测 多目标 可视目标跟踪 聚类分析

Title Tracking-Learning-Detection   Algorithm  And Its Application to  Object Tracking in Video

Abstract

Visual Tracking is an important issue in the field of computer vision, and has great significance in the field of traffic, navigation, and  military and so on. In 2009, Kalal et al proposed an efficient tracking algorithm TLD (Tracking- Learning- Detection) algorithm. TLD, based on tracking, detection and learning, is a long term tracking algorithm for unknown objects. TLD integrates detector, tracker and  learning component together well, and has good robustness. TLD uses a tracker  called LK tracker. This tracker predicts objects based on the successfully tracked points, but the points cannot represent objects well when objects change the appearance or occluded. To solve the problem, we make a cluster analysis on the points the tracker has successfully tracked, and then we pick up the points that can represent the object displacement more accurately. Also, we implement the multi-object tracking algorithm. The improved algorithm maintains the efficiency of TLD, and to some extent, improves the accuracy of it. At the end of the paper, we  make an analysis of the two  algorithms,

and then the deficiencies of the improved algorithm and a follow-up study is put forward．

Keywords     TLD;  multi-Object;  object tracking;   clustering analysis

目次

1  绪论 1

1.1 研究的目的和意义 1

1.2 研究现状 1

1.3 课题任务要求 1

1.4 后续章的组织结构 2

2  相关原理 3

2.1 图像金字塔 3

2.2 Lucas-Kanade 光流法 3

2.3 Forward-Backward 误差 5

2.4 积分图像 6

2.5 聚类分析 7

3   TLD 算法简介 9

3.1 介绍 9

3.2 跟踪器 10

3.3 级联分类器