中心像素加权的无锚框孪生网络跟踪器  

Center pixel weighted anchor-free siamese network tracker

作　　者：谭敏 闫胜业[1] TAN Min;YAN Sheng-ye(School of Automation,Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学自动化学院,江苏南京210044

出　　处：《计算机工程与设计》2023年第7期2047-2053,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61300163)。

摘　　要：为解决SiamRPN++超参数多,对相似背景干扰的判别性不强等问题,提出一种中心像素加权的无锚框孪生网络跟踪器(CPW-Siam)。引入无锚框回归方式,直接在像素点上预测目标框,避免超参数过多的影响;对像素点进行更精确的样本划分,使正样本像素点包含的背景更少,提升特征判别能力;通过中心加权采样,体现正样本像素点不同的关注度,使跟踪器能够做出更精确预测。在VOT2018数据集上的实验结果表明,CPW-Siam的EAO相较SiamRPN++提高了4.5个百分点。To solve the problem that SiamRPN++has many hyper-parameters and similar background distractors,a tracking algorithm based on center pixel weighted anchor-free siamese tracker was proposed(CPW-Siam).The anchor-free regression method was introduced to predict the target bounding boxs directly on pixels,the influence of hyper-parameters was avoided.Pixels were divided more precise,which made positive sample pixels contain less background and enhanced the feature discrimination ability.The focus of different positive sample pixels was reflected using the method of center weighted sampling,showing the different attentions of the positive pixels,so that the tracker could make more accurate prediction.Experimental results on VOT2018 benchmark show that the EAO of CPW-Siam is 4.5%higher than that of SiamRPN++.

关 键 词：超参数 背景干扰 孪生网络 跟踪器 无锚框回归 中心像素 中心加权采样 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]