计算机工程与应用2025，Vol.61Issue(19)：43-59,17.DOI:10.3778/j.issn.1002-8331.2502-0144
基于深度学习的RGBT目标跟踪研究进展
Research Progress of RGBT Object Tracking Based on Deep Learning
摘要
Abstract
Object tracking is a crucial task in the field of computer vision,where single object tracking refers to continu-ously tracking a single target in video sequence.However,visible images depend on lighting conditions,and solely relying on it is no longer sufficient to address various challenges in complex scenes such as low illumination,rainy and foggy.RGBT(RGB-thermal)object tracking refers to the process of combining thermal infrared and visible image data,utilizing the complementary advantages of both modals to jointly achieve tracking task,to improve the robustness and accuracy of object tracking.With the development of deep learning,there are many research achievements in this field,but most of the existing surveys lack an introduction and summary of the frontier research on emerging multi-modal fusion in recent years.This survey first introduces the concept and challenges of RGBT object tracking,and then categorizes existing algo-rithms into five categories for organization and analysis.Following this,it summarizes the current mainstream RGBT object tracking datasets and evaluation indicators,along with a performance comparison of various algorithms on mainstream datasets for researchers to refer to.Finally,it explores the urgent problems and potential directions for RGBT object tracking,to promote further development in the field of tracking.关键词
计算机视觉/深度学习/目标跟踪/热红外图像/多模态融合Key words
computer vision/deep learning/object tracking/thermal infrared image/multi-modal fusion分类
信息技术与安全科学引用本文复制引用
张大伟,王炫,何小卫,郑忠龙..基于深度学习的RGBT目标跟踪研究进展[J].计算机工程与应用,2025,61(19):43-59,17.基金项目
国家自然科学基金(62402449,62272419,62473338) （62402449,62272419,62473338）
浙江省自然科学基金(LQ23F020010) （LQ23F020010）
多模态认知计算安徽省重点实验室(安徽大学)开放基金(MMC202409). （安徽大学）
