An Overview of Correlation-Filter and Deep-Learning Based Single Object Tracking

Abstract

Single object tracking is the key technology in computer vision. It can search, extract, correlate, match, represent and predict the characteristic information of the target in the sequence. It has irreplaceable value in modern military fields such as video retrieval, intelligent monitoring, intelligent interaction, automatic driving, navigation guidance and so on. In 2010 and 2012, correlation-filter and deep-learning technologies were introduced into visual tracking respectively. Since then, both of them have gradually developed into the mainstream. Taking single object tracking as the main task, this paper introduces the principle, process and difficulties of correlation filtering and deep learning tracking technologies, summarizes the classical single target tracking methods based on the above two technologies in recent years, and summarizes and analyzes their basic implementation principles, advantages and disadvantages. The development trend and optimization direction of tracking algorithms in the future are considered and prospected.