基于深度特征和相关滤波器的视觉目标跟踪  

Visual Object Tracking Based on Deep Features and Correlation Filter

作　　者：侯建华[1] 邓雨 陈思萌 项俊[1] Hou Jianhua;Deng Yu;Chen Simeng;Xiang Jun(College of Electronic Information Engineering,South-Central University for Nationalities,Wuhan 43007)

机构地区：[1]中南民族大学电子信息工程学院,武汉30074

出　　处：《中南民族大学学报（自然科学版）》2018年第2期67-73,共7页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家自然科学基金资助项目(61671484;61701548);中南民族大学中央高校基本科研业务费专项资金(CZY18046;CZZ18001)

摘　　要：针对传统的相关滤波方法采用手工特征提取训练样本,限制了跟踪性能的进一步提升,在相关滤波框架下,探讨了深度神经网络VGG-16不同卷积层特征的目标跟踪效果,研究发现,从VGG-16提取的特征相对于传统手工特征具有显著优势;而在深度特征中,以第一层和第五层对目标跟踪性能的提升最为明显.以此为依据,提出了用第一层和第五层特征分别训练相关滤波器、将两者的相关响应图加权后进行目标定位.在OTB2013数据集上的实验结果表明:该方法对跟踪精度和鲁棒性均有进一步的改善.Conventional correlation filter methods employ hand-crafted features to extract training samples,which limits the further promotion of tracking performance. Under the framework of correlation filter,this paper has investigated tracking effect of features from different convolutional layers in deep neural network VGG-16. It is shown in our research that deep features extracted from VGG-16 have a significant advantage compared to conventional hand-crafted ones,and the best results are obtained using features in the first and fifth layers. Based on the above observation,we proposed to train correlation filters separately in the first and fifth layers,and localize the target precisely by weighting response maps produced by the both correlation filters. The experimental results on the OTB2013 dataset demonstrate that the proposed method has improved the tracking accuracy and robustness.

关 键 词：目标跟踪 相关滤波器 响应图 目标定位 深度特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]