基于深度神经网络的目标跟踪算法综述
A Review of Target Tracking Algorithm Based on Deep Neural Network
郭凡 1卢铉宇 1李嘉怡 1王红梅1
作者信息
- 1. 西北工业大学 航天学院,西安 710072
- 折叠
摘要
目标跟踪是根据视频序列中目标的前续信息,对目标的当前状态进行预测.深度学习在目标跟踪领域逐渐广泛应用,本文阐述了目标跟踪算法和深度学习的发展背景,对传统目标跟踪进行了回顾,根据不同的网络任务功能,将基于深度学习的目标跟踪算法分为:基于分类的深度学习目标跟踪算法、基于回归的深度学习目标跟踪算法、基于回归与分类结合的目标跟踪算法,并选取了具有代表性的目标跟踪算法进行实验,对比不同算法之间的特点;最后对目前基于深度学习的目标跟踪方法存在的问题进行分析,对未来发展方向进行展望.实验结果证明,深度孪生跟踪网络在精度与速度上均占优,成为当前主流的跟踪算法框架.
Abstract
Target tracking is to predict the state of the current target according to the target context information of video sequence.Deep learning is gradually widely used in the field of target tracking.This paper elaborates on the de-velopment background of target tracking algorithms and deep learning,and reviews traditional target tracking.Based on different network task functions,target tracking algorithms based deep learning is divided into:deep learning target tracking algorithm based classification,deep learning target tracking algorithm based regression,and target tracking al-gorithm based on the combination of regression and classification.This paper selects representative target tracking algo-rithm for experiments,compares the characteristics of different algorithms.Finally,this paper analyzes the problems of current target tracking algorithms based on deep learning,and prospects the future development direction.The experi-mental results show that deep Siamese tracking networks are superior in accuracy and speed,which is becoming the ma-instream tracking algorithm framework at present.
关键词
目标跟踪/深度学习/神经网络/卷积神经网络/孪生神经网络/生成对抗网络Key words
target tracking/deep learning/neural network/convolutional neural network/Siamese network/generative adversarial network引用本文复制引用
基金项目
国家自然科学基金(62376224)
陕西省重点研发计划(2023-YBGY-232)
出版年
2024
国家科技期刊平台
NSTL
万方数据