Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (21): 24-40.DOI: 10.3778/j.issn.1002-8331.2107-0337

Previous Articles     Next Articles

Review on Object Tracking Methods for Restricted Computing Resources

WU Zhewei, ZHOU Shijie, LIU Qihe   

  1. School of Information and Software Engineering, University of Electronic Science and Technology of China, Chengdu  610000, China
  • Online:2021-11-01 Published:2021-11-04

运算资源受限环境下的目标跟踪算法综述

武哲纬,周世杰,刘启和   

  1. 电子科技大学 信息与软件工程学院,成都 610000

Abstract:

Since the emergence of Deep Neural Network(DNN), the development of target tracking technology has also made great progress. Most of the current research in the field of target tracking is focused on improving the accuracy and efficiency of the algorithm in a computing environment with sufficient computing power. There are relatively few target tracking algorithms in a resource-constrained environment. Therefore, it is essential to develop a tracking network that is still effective in resource-constrained environments. This article systematically sorts out the progress and design concepts of target tracking technology in resource-constrained environments in recent years. This paper introduces the overall workflow of the target tracking task, and makes a summary based on the focus of each tracking method. This paper also summarizes the existing data set in the target tracking task and the evaluation indicators that can be used for model evaluation to facilitate research. The personnel determine the specific data set to be used according to the needs of the actual task. Existing work is combined to explore future research directions.

Key words: object tracking, deep learning, restricted resource

摘要:

自深度神经网络出现以来,目标跟踪技术领域的发展也取得了长足的进步。当前目标跟踪领域的研究大多数都集中在算力充沛的计算环境下提升算法的准确度与效率,在运算资源受限环境下的目标跟踪算法研究相对较少。因此,开发在运算资源受限环境下仍然有效的跟踪网络至关重要。对近年目标跟踪技术所取得的进展与设计理念进行了系统性的梳理,并总结其在适配运算资源受限环境下的改进。介绍了目标跟踪任务的整体工作流程,并根据各跟踪方法的侧重点做出归纳。总结了目标跟踪任务中已有的数据集与可用于模型评估的评价指标,以便于研究人员根据实际任务的需求来确定具体使用的数据集;结合现有的工作,发掘未来的研究方向。

关键词: 目标跟踪, 深度学习, 资源受限