首页 | 本学科首页   官方微博 | 高级检索  
     

特征融合与训练加速的高效目标跟踪
引用本文:刘云,钱美伊,李辉,王传旭.特征融合与训练加速的高效目标跟踪[J].计算机工程与应用,2021,57(10):101-109.
作者姓名:刘云  钱美伊  李辉  王传旭
作者单位:青岛科技大学 信息科学技术学院,山东 青岛 266000
摘    要:基于孪生网络的目标跟踪,存在特征信息欠丰富,跟踪效率有待提高,大型数据集上训练时间长等问题。针对上述问题,提出特征融合与训练加速的高效目标跟踪。增加主干网络参考特征层级,减小下采样,融合多层级参考特征图,提取目标更深度、丰富的语义信息。深度互相关操作得到候选窗口响应(Response of Candidate Windows,RoWs),在其中构建区域建议网络(Region Proposal Network,RPN),通过权衡正负锚点的数量比,使孪生网络性能更加高效、稳定。大型数据集训练孪生网络时,使用均匀滑动漂移采样,代替随机漂移采样算法,在抑制中心偏置现象的同时,显著加快了孪生网络的训练速度。跟踪基准VOT2018上的评估实验结果表明,与所有参考的主流目标跟踪算法相比,所提算法具有最佳的跟踪性能。

关 键 词:目标跟踪  孪生网络  参考特征图融合  深度互相关  候选窗口响应(RoWs)  权衡锚点  漂移采样  训练加速  

Efficient Object Tracking with Feature Fusion and Training Acceleration
LIU Yun,QIAN Meiyi,LI Hui,WANG Chuanxu.Efficient Object Tracking with Feature Fusion and Training Acceleration[J].Computer Engineering and Applications,2021,57(10):101-109.
Authors:LIU Yun  QIAN Meiyi  LI Hui  WANG Chuanxu
Affiliation:School of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong 266000, China
Abstract:There are problems in object tracking based on siamese networks, such as insufficient feature information, tracking efficiency needing to be improved, and long training time on large datasets. In order to solve the above problems, efficient object tracking based on feature fusion and training acceleration is proposed. The reference feature level of backbone network is increased, down-sampling is reduced, and multi-level reference feature maps are fused with the purpose of extracting deeper and more abundant semantic targets information. Depth-wise cross correlation is used to get RoWs (Response of Candidate Windows), in which the Region Proposal Network(RPN) is constructed. By balancing the number of positive and negative anchors, the performance of siamese networks is more efficient and steady. When training siamese networks on large datasets, uniformly sliding shift sampling is used instead of randomly shift sampling, which significantly accelerates training speed at the time of suppressing center bias phenomenon. Experimental evaluation results on tracking benchmark VOT2018 show that the proposed algorithm has the best tracking performance compared with all reference mainstream object tracking algorithms.
Keywords:object tracking  siamese network  reference feature map fusion  depth-wise cross correlation  Response of Candidate Windows(RoWs)  balancing anchors  shift sampling  training acceleration  
本文献已被 万方数据 等数据库收录!
点击此处可从《计算机工程与应用》浏览原始摘要信息
点击此处可从《计算机工程与应用》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号