摘要
视觉跟踪是一个重要的计算机视觉任务,有着广泛的应用,由于现实场景中存在着众多困难,视觉跟踪仍是一个活跃的研究领域。判别式分类器是现代跟踪方法中的一个核心组成部分,其在线学习一个二值分类器以在每一帧中区分目标与背景,充分利用机器学习中丰富的学习算法,取得了许多突破。相关滤波器已成功应用到目标检测和识别中,其由于计算效率高,近年来作为一种判别式跟踪方法被应用到视觉跟踪领域,取得了很好的效果。首先简要介绍了判别式跟踪算法;然后对相关滤波器基本理论及几种典型的相关滤波器构造方法进行了描述;最后重点介绍了近年来相关滤波器在视觉跟踪中的应用及研究进展,并总结了可能的研究方向和发展趋势。
Visual object tracking is a fundamental task in many computer vision applications which is still an active research field due to the challenges in real scenes.The core component of most modern trackers is a discriminative classifier which can use the abundant algorithms in machine learning to learn a binary classifier online to separate the object from the surrounding environment.Due to the high computational efficiency,as a discriminative tracking method,correlation filters which have been successfully applied to a variety of pattern recognition applications are introduced to the topic of visual tracking in recent years.The discriminative learning methods in visual tracking were introduced briefly firstly.Then the fundamental theory and some kinds of the typical methods of correlation filters were described.Finally,a detailed review of the applications of the correlation filters in visual tracking was provided,and the future applications and research trends were discussed.
出处
《计算机科学》
CSCD
北大核心
2016年第11期1-5,18,共6页
Computer Science
关键词
视觉跟踪
判别式学习方法
相关滤波器
Visual tracking
Discriminative learning
Correlation filter