特征增强和双模板更新的目标跟踪算法  

Feature Enhancement and Dual-Template UpdatingStrategy for Object Tracking Algorithm

作　　者：符强[1,2,4] 梁栩欣 纪元法 任风华[3] FU Qiang;LIANG Xuxin;JI Yuanfa;REN Fenghua(Guilin University of Electronic Technology,Guangxi Key Laboratory of Precision Navigation Technology and Application,Guilin 541000,China;Guilin University of Electronic Technology,College of Information and Communication,Guilin 541000,China;Guilin University of Electronic Technology,College of Electrical Engineering and Automation,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]桂林电子科技大学电子工程与自动化学院,广西桂林541000 [4]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》2023年第10期7-12,共6页Electronics Optics & Control

基　　金：国家自然科学基金(61561016,62061010);广西科技厅项目(桂科AA19182007,桂科AA19254029)。

摘　　要：为改进在复杂场景下的跟踪性能,提出了一种特征增强和双模板更新的目标跟踪算法。首先,提出了一种改进的特征提取网络,并将深层特征和浅层特征融合,再利用通道空间注意力模块对该融合特征进行强化,获得表征能力增强的特征。其次,提出了一种双模板更新策略,将近邻高置信度图像帧保留为备份模板,当跟踪响应图置信度较低时,将初始模板与备份模板进行加权融合得到新的模板,再重新进行跟踪预测。最后,在数据集OTB-100和VOT-2017上进行跟踪性能评估,实验结果表明,所提算法提升了在遮挡、光照变化、背景杂乱等复杂场景下的跟踪成功率和跟踪准确率。To improve the tracking performance in complex scenarios,an object tracking algorithm with feature enhancement and dual-template updating is proposed.Firstly,an improved feature extraction network is proposed,and deep-layer features and shallow-layer features are fused,and then the spatial-channel attention module is utilized to enhance the fused features,so that better capability of feature representation is obtained.Secondly,a dual-template updating strategy is proposed,and the adjacent high-confidence frame of the image is reserved as the backup template,which is utilized to implement weighted fusion with the initial template to obtain new templates for conducting tracking prediction again when the confidence of the tracking response map is low.Finally,the tracking performance is evaluated on the datasets of OTB-100 and VOT-2017,and the experimental results show that the proposed algorithm improves the tracking success rate and accuracy in complex scenarios such as occlusion,illumination change and background clutter.

关 键 词：目标跟踪 特征融合 注意力模块 模板更新 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]