注意力机制的多尺度单目标跟踪算法

doi:10.19665/j.issn1001-2400.2021.05.014

西安电子科技大学学报 ›› 2021, Vol. 48 ›› Issue (5): 110-116.doi: 10.19665/j.issn1001-2400.2021.05.014

注意力机制的多尺度单目标跟踪算法

宋建锋(),苗启广(),王崇晓(),徐浩(),杨瑾()

西安电子科技大学计算机科学与技术学院,陕西西安 710071

收稿日期:2021-05-10 出版日期:2021-10-20 发布日期:2021-11-09
通讯作者: 苗启广
作者简介:宋建锋(1978—),男,讲师,博士,E-mail: jfsong@mail.xidian.edu.cn|王崇晓(1996—),男,西安电子科技大学硕士研究生,E-mail: 33760709@qq.com|徐浩(1998—),男,西安电子科技大学硕士研究生,E-mail: haoxu_mail@163.com|杨瑾(1998—),女,西安电子科技大学硕士研究生,E-mail: yangjin980110@163.com
基金资助:
国家重点研发计划(2019YFC0000238)

Multi-scale single object tracking based on the attention mechanism

SONG Jianfeng(),MIAO Qiguang(),WANG Chongxiao(),XU Hao(),YANG Jin()

School of Computer Science and Technology,Xidian University,Xi’an 710071,China

Received:2021-05-10 Online:2021-10-20 Published:2021-11-09
Contact: Qiguang MIAO

摘要/Abstract

摘要：

在单目标跟踪过程中,由于存在目标遮挡、目标消失、相似目标干扰等问题,导致算法错误跟踪目标,跟踪精度下降,并且错误的结果将会参与到模型更新中,使得跟踪精度进一步下降。针对这一问题,提出了基于注意力机制的多尺度单目标跟踪算法。该算法使用Inception网络非对称卷积思想,在增加多尺度卷积核的同时减少参数量,非对称卷积可以有效地结合局部特征和全局特征,提高跟踪的鲁棒性。在模型参数更新阶段,采用基于注意力机制的网络在线更新算法,结合每一帧的结果响应图和注意力响应图计算得到该帧的跟踪结果得分,从而在模型更新时剔除不包含目标的视频帧,强化了网络对目标和背景的判别能力,使网络能够快速学习到目标的外观变化,提高算法对目标的跟踪能力。在实验部分,将该算法在OTB-100数据集与其他先进的跟踪算法进行对比,在准确率和成功率方面相较ATOM算法分别有0.9%和0.8%的提升,提升精度的同时也更容易找回丢失的目标。

关键词: 深度学习, 单目标跟踪, 注意力机制, 卷积神经网络

Abstract:

In the process of single object tracking,due to problems of occlusion,disappearance and similar target interference,the tracking accuracy of the algorithm will be reduced.In order to solve these problems,a multi-scale single target tracking algorism based on the attention mechanism is proposed which uses asymmetric convolution to extract the multi-scale feature while reducing the parameters.It combines local features and global features to improve tracking robustness.The online update algorithm based on the attention mechanism is used which combines the response diagram and attention diagram to calculate a score which is used to weed out frames without targets.The attention mechanism strengthens the ability to distinguish the target and background,makes the network quickly adapt to the changes of the appearance of the targets,and improves the tracking performance of the algorithm.The algorithm is tested on OTB-100 datasets with other advanced tracking methods.Compared to the ATOM,the accuracy and success rate of our method are improved by 0.9% and 0.8% respectively and it is easier to retrieve the target after it is lost.

Key words: deep learning, single object tracking, attention, convolutional neural network

中图分类号:

TP37

宋建锋,苗启广,王崇晓,徐浩,杨瑾. 注意力机制的多尺度单目标跟踪算法[J]. 西安电子科技大学学报, 2021, 48(5): 110-116.

SONG Jianfeng,MIAO Qiguang,WANG Chongxiao,XU Hao,YANG Jin. Multi-scale single object tracking based on the attention mechanism[J]. Journal of Xidian University, 2021, 48(5): 110-116.

图/表 7

图1

图2

图3

图4

图5

图6

图7

参考文献 17

[1]	孟琭, 杨旭. 目标跟踪算法综述[J]. 自动化学报, 2019, 45(7):1244-1260.
	MENG Lu, YANG Xu. A Survey of Object Tracking Algorithms[J]. Acta Automatica Sinica, 2019, 45(7):1244-1260.
[2]	孔晓东. 智能视频监控技术研究[D]. 上海:上海交通大学, 2008.
[3]	MA C, HUANG J B, YANG X K, et al. Hierarchical Convolutional Features for Visual Tracking[C]// IEEE International Conference on Computer Vision.Piscataway:IEEE, 2015:3074-3082.
[4]	BERTINETTO L, VALMADRE J, HENRIQUES J K, et al. Fully-Convolutional Siamese Networks for Object Tracking[C]// European Conference on Computer Vision.Heidelberg:Springer, 2016:850-865.
[5]	NAM H, HAN B. Learning Multi-Domain Convolutional Neural Networks for Visual Tracking[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2016:4293-4302.
[6]	DANELLJAN M, BHAT G, KHAN F S, et al. ATOM:Accurate Tracking by Overlap Maximization[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2019:4660-4669.
[7]	成磊, 王玥, 田春娜. 一种添加残差注意力机制的视觉目标跟踪算法[J]. 西安电子科技大学学报, 2020, 47(6):151-160.
	CHENG Lei, WANG Yue, TIAN Chunna. Residual Attention Mechanism for Visual Tracking[J]. Journal of Xidian University, 2020, 47(6):151-160.
[8]	SZEGEDY C, LIU W, JIA Y, et al. Going Deeper with Convolutions[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2015:1-9.
[9]	BOLME D S, BEVERIDGE J R, DRAPER B A, et al. Visual Object Tracking Using Adaptive Correlation Filters[C]// 23rd IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2010:2544-2550.
[10]	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the Inception Architecture for Computer Vision[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2016:2818-2826.
[11]	GUO D, SHAO Y, CUI Y, et al. Graph attention tracking[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2021:9543-9552.
[12]	PU S, SONG Y, MA C, et al. Deep Attentive Tracking Via Reciprocative Learning (2018)[J/OL]. [2018-10-15]https://arxiv.org/pdf/1810.03851/pdf.
[13]	HENRIQUES J F, CASEIRO R, MARTINS P, et al. High-Speed Tracking with Kernelized Correlation Filters[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 37(3):583-596.
[14]	HE K, ZHANG X Y, REN S Q, et al. Deep Residual Learning for Image Recognition[C]// IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2016:770-778.
[15]	WU Y, LIM J, YANG M H. Online Object Tracking:A Benchmark[C]// 2013 IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2013:2411-2418.
[16]	SONG Y, MA C, GONG L, et al. Crest:Convolutional Residual Learning for Visual Tracking[C]// Proceedings of the IEEE International Conference on Computer Vision.Piscataway:IEEE, 2017:2555-2564.
[17]	BERTINETTO L, VALMADRE J, HENRIQUES J F, et al. Fully-Convolutional Siamese Networks for Object Tracking[C]// European Conference on Computer Vision.Heidelberg:Springer, 2016:850-865.

注意力机制的多尺度单目标跟踪算法

Multi-scale single object tracking based on the attention mechanism

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 17

相关文章 15

Metrics

本文评价

推荐阅读 10

[1]	李源,崔玉爽,王伟. 一种基于字词双通道网络的文本情感分析方法[J]. 西安电子科技大学学报, 2021, 48(6): 179-186.
[2]	刘佳玮,张文辉,寇晓丽,李雁妮. 增强型深度对抗样本攻击防御算法[J]. 西安电子科技大学学报, 2021, 48(6): 23-31.
[3]	于浩洋,尹良,李书芳,吕顺. 生成对抗网络小样本雷达调制信号识别算法[J]. 西安电子科技大学学报, 2021, 48(6): 96-104.
[4]	董如婵,焦李成,赵进,沈维燕. 一种深度融合机制的遥感图像目标检测技术[J]. 西安电子科技大学学报, 2021, 48(5): 128-138.
[5]	李鹏,冯存前,许旭光,唐子翔. 一种利用贝叶斯优化的弹道目标微动分类网络[J]. 西安电子科技大学学报, 2021, 48(5): 139-148.
[6]	孙彦景,魏力,张年龙,云霄,董锴文,葛敏,程小舟,侯晓峰. 联合DD-GAN和全局特征的井下人员重识别方法[J]. 西安电子科技大学学报, 2021, 48(5): 201-211.
[7]	闫佳,曹玉东,任佳兴,陈冬昊,李晓会. 深度非对称压缩型哈希算法[J]. 西安电子科技大学学报, 2021, 48(5): 212-221.
[8]	宁阳,杜建超,韩硕,杨传凯. 改进DeeplabV3+的火焰分割与火情分析方法[J]. 西安电子科技大学学报, 2021, 48(5): 38-46.
[9]	周鹏,杨军. 采用神经网络架构搜索的遥感影像分割方法[J]. 西安电子科技大学学报, 2021, 48(5): 47-57.
[10]	戚艳军,孔月萍,王佳婧,朱旭东. 一种LSTM与CNN相结合的步态识别方法[J]. 西安电子科技大学学报, 2021, 48(5): 78-85.
[11]	杨云航,闵连权. 采用空洞卷积的多尺度融合草图识别模型[J]. 西安电子科技大学学报, 2021, 48(5): 92-99.
[12]	张宇浩,程培涛,张书豪,王秀美. 一种自适应权重学习的轻量超分辨率重建网络[J]. 西安电子科技大学学报, 2021, 48(5): 15-22.
[13]	陈昌川,王海宁,黄炼,黄涛,李连杰,黄向康,代少升. 一种基于局部表征的面部表情识别算法[J]. 西安电子科技大学学报, 2021, 48(5): 100-109.
[14]	回海生,张雪英,吴泽林,李凤莲. 一种主辅路径注意力补偿的脑卒中病灶分割方法[J]. 西安电子科技大学学报, 2021, 48(4): 200-208.
[15]	宋剑桥,王峰,牛锦,师泽洲,马军辉. 一种面向时空神经网络的潜在情绪识别方法[J]. 西安电子科技大学学报, 2021, 48(4): 159-167.