ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2015, Vol. 52 ›› Issue (1): 177-190.doi: 10.7544/issn1000-1239.2015.20130995

• 人工智能 • 上一篇    下一篇

基于外观模型学习的视频目标跟踪方法综述

张焕龙1,2,胡士强1,杨国胜3,   

  1. 1(上海交通大学航空航天学院 上海 200240); 2(洛阳理工学院计算机与信息工程系 河南洛阳 471023); 3(中央民族大学信息工程学院 北京 100081) (zhl_lit@sjtu.edu.cn)
  • 出版日期: 2015-01-01
  • 基金资助: 
    基金项目:国家自然科学基金项目(610741006,61374161)|河南省科技厅科技攻关项目(132102210513)

Video Object Tracking Based on Appearance Models Learning

Zhang Huanlong1,2,Hu Shiqiang1,Yang Guosheng3   

  1. 1(School of Aeronautics and Astronautics,Shanghai Jiao Tong University, Shanghai 200240); 2(Department of Computer and Information Engineering, Luoyang Institute of Scicnce and Technology, Luoyang, Henan 471023); 2(School of Information Engineering, Minzu University of China, Beijing 100081)
  • Online: 2015-01-01

摘要: 视频跟踪是机器视觉领域中的热点研究问题,在过去的几十年内得到了广泛研究.为了获得鲁棒的跟踪效果,设计能够适应跟踪目标外观变化的外观模型成为算法研究中的一种重要内容.近年来,将机器学习理论引入外观模型设计中的思想大大推动了视频跟踪研究的发展.为了使读者能够快速了解其发展的趋势并且掌握基于外观模型学习跟踪算法研究的技术,在介绍外观模型学习跟踪算法原理和机制的基础上,重点综述了外观模型学习跟踪方法的研究进展,包括目标特征描述和3类主要目标外观建模方式及其各自研究过程中跟踪方法的对比与分析,进而总结了外观模型学习跟踪算法在理论及应用方面的研究现状,最后提出进一步研究的主要发展内容和趋势.

关键词: 外观模型学习, 稀疏表示, 度量学习, 视频跟踪, 半监督学习

Abstract: Visual tracking is an active reasch topic in the field of computer vision and has been well studied in the last decades. A key component for achieving robust tracking is the tracker’s capability of updating its internal representation of tragets to capture the varying appearance. Although numberous approaches have been proposed, many challenging problems still remain in designing an effective model of the appearance of tracked objects. In recent years, the methods of appearance model associated with statistical learning have been promoting the study for video object tracking. To help reader swiftly learn the rencent advances and trends so as to easily grasp the key problems of visual object tracking based on appearance models learning, a detailed review of the existing appearance learning models is provided. Here, the mechanism of the tracking algorithm based on appearance model learning is introduced firstly. Then the state-of-the-art feature descriptors are analyzed to show their different performance. Meanwhile, the tracking progress is categorized into three main groups, and the character of representative methods in each group are compared and analyzed in detail. Finally, the current research on the tracking methods based appearance model learning is summarized and classified, and the further application and research trend is discussed.

Key words: appearance model learning, sparse representation, metric learning, video tracking, semi-supervised learning

中图分类号: