RGBT tracking via reliable feature configuration  

作　　者：Zhengzheng TU Wenli PAN Yunsheng DUAN Jin TANG Chenglong LI 

机构地区：[1]Anhui Provincial Key Laboratory of Multimodal Cognitive Computation,Hefei 230601,China [2]School of Computer Science and Technology,Anhui University,Hefei 230601,China [3]Network Information Center,Anhui University,Hefei 230601,China

出　　处：《Science China(Information Sciences)》2022年第4期124-136,共13页中国科学（信息科学）（英文版）

基　　金：supported by Natural Science Foundation of Anhui Higher Education Institution of China(Grant Nos.KJ2020A0033,KJ2019A0005,KJ2019A0026);Major Project for New Generation of AI(Grant No.2018AAA0100400);National Natural Science Foundation of China(Grant No.61976003);NSFC Key Projects in International(Regional)Cooperation and Exchanges(Grant No.61860206004).

摘　　要：There is an increasing interest in RGBT tracking recently because of the complementary benefits of RGB and thermal infrared data.However,the reliability of each modality will change over time quite possibly,and the modality with bad reliability will disturb tracking performance.Thus,we propose a novel reliability-based feature configuration approach in the correlation filter framework for robust RGBT tracking.Specifically,we configure a feature set based on RGB,thermal,and RGBT data.To measure the reliabilities of different feature configurations,we equip each feature configuration with a tracker and design a guideline judging whether the tracker is reliable.We use the tracker with the best reliability for tracking.Experimental results show that the proposed tracker achieves promising performance against other RGBT tracking methods.

关 键 词：RGBT tracking MULTI-MODAL reliability guideline feature configuration correlation filter 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]