多视图专家组区域建议预测的视觉跟踪  

Multi-view multi-expert groups region proposal prediction for tracking

作　　者：单彬 丁昕苗[1] 王铭淏 郭文[1] SHAN Bin;DING Xin-miao;WANG Ming-hao;GUO Wen(School of Information and Electronic Engineering,Shandong Technology and Business University,Yantai 264009,China)

机构地区：[1]山东工商学院信息与电子工程学院,山东烟台264009

出　　处：《计算机工程与设计》2024年第2期459-466,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62072286、61876100、61572296);山东省自然科学基金项目(ZR2015FL020)。

摘　　要：为解决大多数最新的目标跟踪器都面临着判别特征表示缺乏多样性、目标定位过于模糊以及正样本的数量要求问题,提出一种基于多视图的顶层特征的区域建议网络的跟踪预测学习算法。融合多种视图的特征表示方式,利用丰富多样的语义信息,有效解决判别特征过于单一的问题。在扩展的边界框上构建多个支持向量机模型并加入区域建议网络模块,精确优化边界框,预测最优的目标位置,缓解目标定位过于模糊和正样本的数量有限的问题。通过大量视频基准序列对方法的综合评价,其结果表明,提出方法融合了轻量化的深度学习模型和多视图专家组的优点,使跟踪性能有了显著提升。To solve the problems that most of the latest target trackers are faced with,such as nondiverse discriminate feature representation,coarse object locator,and limited quantities of positive samples,a tracking prediction learning algorithm based on multi-view multi-expert region proposal prediction algorithm was proposed.Multiple views and exploits powerful multiple sources of information were integrated,which solved nondiverse discriminate feature representation problem effectively.Multiple SVM classifier models were built on the expanded bounding boxes and the regional suggestion network module was added to accurately optimize it to predict optimal object location,which naturally alleviated the coarse object locator and limited quantities of positive samples problems at the same time.A comprehensive evaluation of the proposed approach on various video benchmark sequences was performed.The evaluation results demonstrate that the method proposed can significantly improve the tracking performance by combining the advantages of lightweight deep learning model and multi-view expert groups.

关 键 词：区域建议预测 特征判别机制 多专家组模型 多视图模型 特征融合 视觉跟踪 深度学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]