基于动态权重的双分支孪生网络目标跟踪算法

中国民航大学学报 ›› 2022, Vol. 40 ›› Issue (5): 15-22.

基于动态权重的双分支孪生网络目标跟踪算法

韩萍^a，王皓韡^b，方澄^a

（中国民航大学a.电子信息与自动化学院；b.计算机科学与技术学院，天津300300）

收稿日期:2021-04-12 修回日期:2021-05-18 出版日期:2022-10-15 发布日期:2023-10-27
作者简介:韩萍（1966—），女，天津人，教授，博士，研究方向为信号与信息处理、SAR 目标检测与识别等.
基金资助:
中国民航大学科研启动基金项目（2017QD05S）；中央高校基本科研业务费专项（3122018C005）

Object tracking algorithm based on dynamic weight in dual branch siamese network

HAN Ping^a , WANG Haowei^b , FANG Cheng^a

（a. College of Electronic Information and Automation; b. College of Computer Science and Technology, CAUC, Tianjin 300300, China）

Received:2021-04-12 Revised:2021-05-18 Online:2022-10-15 Published:2023-10-27

摘要/Abstract

摘要： 以基于全卷积孪生网络的目标跟踪（SiamFC，fully-convolutionalsiamesenetworksforobjecttracking）算法为代表的部分深度孪生网络目标跟踪算法均是针对目标外观信息进行设计的，易受高速移动、运动模糊、光照变化等因素的影响，造成跟踪目标漂移或丢失。为了提高算法对目标外观变化的适应能力，给出一种基于动态权重的双分支孪生网络目标跟踪算法，以替换特征提取网络后的SiamFC算法作为外观分支，在此基础上增加利用双重注意力强化信息提取的语义分支作为外观分支的有效补充。跟踪阶段利用动态权重系数结合两分支的跟踪结果，有效抑制了目标外观变化对跟踪算法的影响，提升了算法的跟踪精度和鲁棒性。在4个标准目标跟踪数据集OTB2015、UAV20L、UAV123和GOT-10k上验证了本文算法的有效性，平均跟踪帧率为47帧/s，满足跟踪实时性要求。

关键词: 视频目标跟踪, 孪生网络, 注意力机制, 动态权重

Abstract: Some of siamese tracking algorithms represented by SiamFC are designed for target appearance information.However, appearance information is easily affected by factors such as high-speed movement, motion blur, and illumination changes, resulting in tracking drift or target loss. In order to improve the ability of the algorithm to adapt target appearance changes, this paper adopts a dual -branch siamese network tracking algorithm based on dynamic weights. Based on improved SiamFC algorithm as the appearance branch, a semantic branch using dual attention mechanism to enhance information extraction is added as an effective supplement. In the tracking stage, a dynamic weight is used to fuse the tracking results of the two branches, which effectively suppresses the influence of target appearance changes, also improves the tracking accuracy and robustness of the algorithm. Our algorithm has been validated on four standard object tracking data sets (OTB2015, UAV123, UAV20L, GOT-10k). The average tracking frame rate is 47 frames/s, which meets the real-time tracking requirements.

Key words: video object tracking, siamese network, attention mechanism, dynamic weight

中图分类号:

TP391.41

韩萍, 王皓韡, 方澄. 基于动态权重的双分支孪生网络目标跟踪算法[J]. 中国民航大学学报, 2022, 40(5): 15-22.

HAN Ping , WANG Haowei , FANG Cheng. Object tracking algorithm based on dynamic weight in dual branch siamese network[J]. Journal of Civil Aviation University of China, 2022, 40(5): 15-22.

参考文献 13

[1]	DANELLJAN M, ROBINSON A, SHAHBAZ KHAN F, et al． Beyond correlation filters: learning continuous convolution operators for visual tracking[C]//Computer Vision-ECCV 2016, 2016: 472-488．
[2]	DANELLJAN M, BHAT G, KHAN F S, et al． ECO: efficient convolution operators for tracking[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition, July 21-26, 2017, Honolulu, HI, USA． IEEE, 2017: 6931-6939．
[3]	BERTINETTO L, VALMADRE J, HENRIQUES J F, et al． Fully-con－ volutional Siamese networks for object tracking [C]//European Confer－ ence on Computer Vision 2016, Amsterdam, The Netherlands, October 8-10 and 15-16, 2016: 850-865．
[4]	VALMADRE J, BERTINETTO L, HENRIQUES J, et al． End-to-end representation learning for correlation filter based tracking [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition, July 2126, 2017, Honolulu, HI, USA． IEEE, 2017: 5000-5008．
[5]	LI B, YAN J J, WU W, et al． High performance visual tracking with Siamese region proposal network[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 18-23, 2018, Salt Lake City, UT, USA． IEEE, 2018: 8971-8980．
[6]	WANG N, SONG Y B, MA C, et al． Unsupervised deep tracking [C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recogni－ tion(CVPR), June 15-20, 2019, Long Beach, CA, USA． IEEE, 2019: 1308-1317．
[7]	王玲, 王家沛, 王鹏, 等．融合注意力机制的孪生网络目标跟踪算法研究[J]．计算机工程与应用, 2021, 57(8): 169-174．
[8]	WANG Q, TENG Z, XING J L, et al． Learning attentions: residual at－ tentional Siamese network for high performance online visual tracking [C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 18-23, 2018, Salt Lake City, UT, USA． IEEE, 2018: 4854-4863．
[9]	范颖, 宋晓宁．用于目标跟踪的孪生渐进注意引导融合网络[J]．计算机辅助设计与图形学学报, 2021, 33(2): 199-206．
[10]	WU Y, LIM J, YANG M H． Object tracking benchmark[J]． IEEE Trans－ actions on Pattern Analysis and Machine Intelligence, 2015, 37 (9): 1834-1848．
[11]	MUELLER M, SMITH N, GHANEM B． A benchmark and simulator for UAV tracking[C]//Computer Vision-ECCV 2016, 2016: 445-461．
[12]	HUANG L H, ZHAO X, HUANG K Q． GOT-10k: a large high-diversity benchmark for generic object tracking in the wild[J]． IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(5): 1562-1577．
[13]	DENG J, DONG W, SOCHER R, et al． ImageNet: A large-scale hierar－ chical image database[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition, June 20-25, 2009, Miami, FL, USA． IEEE, 2009: 248-255． [14] LI B, WU W, WANG Q, et al． SiamRPN++: evolution of Siamese visual tracking with very deep networks[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 15-20, 2019, Long Beach, CA, USA． IEEE, 2019: 4277-4286.