摘要
视觉目标跟踪是对目标位置、速度、运动轨迹等信息检测与预测的技术。该技术融合了计算机视觉、图像处理、深度学习等众多技术。将对目标跟踪算法发展情况以及研究现状进行梳理,首先介绍目前常用的基准数据集;其次指出生成式算法与判别式算法差异;再对传统的生成式算法进行简单分析总结;随后围绕算法框架分别介绍相关滤波框架、深度学习框架、孪生网络框架、Transformer框架的判别式算法并分析不同算法的优缺点;最后分析目前动态目标跟踪存在的问题并展望。
The position,velocity,motion trajectory and other information of the target are detected and predicted by visual target tracking.Computer vision,image processing,deep learning and many other technologies are integrated in visual target tracking.The development and research status of target tracking algorithm were introduced.Firstly,the commonly used benchmark data sets were introduced.Secondly,the difference between generative algorithm and discriminant algorithm was pointed out.In addition,the traditional generative algorithm was simply analyzed and summarized.Then,the discriminant algorithms of correlation filtering framework,deep learning framework,twin network framework and transformer framework were introduced,and the advantages and disadvantages of different algorithms were analyzed;Finally,the existing problems of dynamic target tracking were analyzed and prospected.
作者
彭建盛
许恒铭
李涛涛
侯雅茹
PENG Jian-sheng;XU Heng-ming;LI Tao-tao;HOU Ya-ru(College of Electrical and Information Engineering,Guangxi University of Science and Technology,Liuzhou 545000,China;Colleges of Artificial Intelligence and Smart Manufacturing,Hechi University,Yizhou 546300,China)
出处
《科学技术与工程》
北大核心
2021年第35期14871-14881,共11页
Science Technology and Engineering
基金
国家自然科学基金(62063006)
广西自然科学基金(2018GXNSFAA281164)。
关键词
目标跟踪
生成式算法
相关滤波
深度学习
孪生网络
target tracking
generative algorithm
correlation filtering
deep learning
siamese network
