摘要
为提升基于Siamese网络视觉跟踪算法的特征表达能力和判别能力,以获得更好的跟踪性能,提出了一种轻量级的基于二阶注意力的Siamese网络视觉跟踪算法。使用轻量级VGG-Net作为Siamese网络的主干,获取目标的深度特征;在Siamese网络的末端并行使用所提残差二阶池化网络和二阶空间注意力网络,获取具有通道相关性的二阶注意力特征和具有空间相关性的二阶注意力特征;使用残差二阶通道注意力特征和二阶空间注意力特征,通过双分支响应策略实现视觉跟踪。利用GOT-10k数据集对所提算法进行端到端的训练,并在OTB100和VOT2018数据集上进行验证。实验结果表明:所提算法的跟踪性能取得了显著提升,与基准算法SiamFC相比,在OTB100数据集上,精度和成功率分别提高了0.100和0.096,在VOT2018数据集上,预期平均重叠率(EAO)提高了0.077,跟踪速度达到了48帧/s。
To improve the feature expression ability and discriminative ability of the visual tracking algorithm based on Siamese network and obtain better tracking performance,a lightweight Siamese network visual tracking algorithm based on second-order attention is proposed.Firstly,to obtain deep features of the object,the lightweight VGG-Net is used as the backbone of the Siamese network.Secondly,the residual second-order pooling network and the second-order spatial attention network are used in parallel at the end of the Siamese network to obtain the secondorder attention features with channel correlation and the second-order attention features with spatial correlation.Finally,visual tracking is achieved through a double branch response strategy using the residual second-order channel attention features and the second-order spatial attention features.The proposed algorithm is trained end-to-end with the GOT-10k dataset and validated on the datasets OTB100 and VOT2018.The experimental results show that the tracking performance of the proposed algorithm has been significantly improved.Compared with the baseline algorithm SiamFC,on dataset OTB100,the precision and the success are increased by 0.100 and 0.096,respectively;on dataset VOT2018,the expected average overlap(EAO)increased by 0.077,tracking speed reached 48 frame/s.
作者
侯志强
陈茂林
马靖媛
郭凡
余旺盛
马素刚
HOU Zhiqiang;CHEN Maolin;MA Jingyuan;GUO Fan;YU Wangsheng;MA Sugang(School of Computer Science and Technology,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;School of Information and Navigation,Air Force Engineering University,Xi’an 710077,China)
出处
《北京航空航天大学学报》
EI
CAS
CSCD
北大核心
2024年第3期739-747,共9页
Journal of Beijing University of Aeronautics and Astronautics
基金
国家自然科学基金(62072370)。
关键词
Siamese网络
视觉跟踪
残差二阶池化网络
二阶空间注意力网络
双分支响应策略
Siamese network
visual tracking
residual second-order pooling network
second-order spatial attention network
double branch response strategy
