在线机器学习跟踪算法的研究进展

Research Progress on Visual Tracking Algorithms Based on Online Machine Learning

分类介绍在线机器学习跟踪算法的研究现状,比较各种算法的优缺点.研究表明:每一种跟踪算法都有其自身的优点和缺点,通常情况下只能处理某一些特定类型的变化,很难确保某一特定类型的跟踪算法能够处理复杂跟踪场景中的所有不确定因素.最后,针对在线学习算法容易产生误差积累,最终发生目标漂移的问题,提出使用多跟踪器的融合,实现鲁棒跟踪等相应的解决方案.

In this paper, the tracking algorithms based on online learning are reviewed, and the advantages and disadvantages of various tracking algorithms are compared. It is found that each kind of tracking method has its strengths and weaknesses and is applicable for handling one or a few types of challenges, it is difficult, if not impossible, for a single tracking method to work under a variety of tracking scenarios. Finally, to address the target drifting problem caused by the error accumulation, we propose a fusion strategy using multiple trackers to achieve robust tracking results.