自适应特征融合的加权响应孪生网络跟踪算法  

A Weighted Response Siamese Network Tracking Algorithm Based on Adaptive Feature Fusion

作　　者：符强[1,2] 谢志安 纪元法[1,2,3] 任风华 FU Qiang;XIE Zhian;JI Yuanfa;REN Fenghua(Guangxi Key Laboratory of Precision Navigation Technology and Application,Guilin University of Electronic Technology,Guilin 541000,China;Information and Communication School,Guilin University of Electronic Technology,Guilin 541000,China;National&Local Joint Engineering Research Center of Satellite Navigation Positioning and Location Service,Guilin 541000,China)

机构地区：[1]桂林电子科技大学广西精密导航技术与应用重点实验室,广西桂林541000 [2]桂林电子科技大学信息与通信学院,广西桂林541000 [3]卫星导航定位与位置服务国家地方联合工程研究中心,广西桂林541000

出　　处：《电光与控制》2024年第9期25-30,共6页Electronics Optics & Control

基　　金：国家自然科学基金(62061010,62161007);广西省科技厅项目(桂科AA20302022,桂科AB21196041)。

摘　　要：针对复杂场景下跟踪器难以提取丰富的特征信息导致出现漂移或者跟踪丢失的问题,提出了一种基于自适应特征融合的加权响应目标跟踪算法。首先,使用改进的VGG16网络来提高判别能力;其次,采用残差语义嵌入模块,将深层语义信息引入浅层特征;然后,将浅层特征响应和深层特征响应进行加权融合,提高定位精度和判别能力。实验结果表明,相比基准算法,所提算法在OTB2015和VOT2017数据集上的跟踪成功率和精度等评价指标均获得提升。A weighted response target tracking algorithm based on adaptive feature fusion is proposed to address the problem of drift or tracking loss due to the difficulty of extracting rich feature information by the tracker in complex scenes.Firstly,an improved VGG16 network is used to improve the discriminative ability.Secondly,a residual semantic embedding module is employed to introduce deep semantic information into shallow features,and then the shallow feature response and deep feature response are weighted and fused to improve the localization accuracy and discriminative ability.The experimental results show that,compared with the benchmark algorithm,the evaluation indexes of the proposed algorithm,such as tracking success rate and accuracy,are both improved on the OTB2015 and VOT2017 datasets.

关 键 词：目标跟踪 孪生网络 注意力机制 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]