跳到主要导航 跳到搜索 跳到主要内容

MSTracker: A Multi-scale Spatio-temporal Framework for UAV Multi-Object Tracking

  • Yan Ding*
  • , Minjin Zhao
  • , Yuchen Ling
  • , Shupeng Guo
  • , Yixiao Fan
  • , Luheng Cui
  • *此作品的通讯作者
  • Beijing Institute of Technology
  • Ltd.

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

Aiming at the challenges of trajectory fragmentation and identity switches caused by similar appearances and frequent occlusions in Unmanned Aerial Vehicle (UAV) aerial videos, this paper proposes a novel Multi-scale Spatio-temporal tracking framework, termed MSTracker. The core of MSTracker consists of three key components: 1) A Multi-Scale Feature Extraction model with a Feature Map Quality Weighting Scheme to enhance target representation; 2) A Spatio-Temporally Guided Trajectory Update Model that leverages long-term context and dynamic trajectory management to reduce identity switches; 3) A Dynamically Weighted Fusion Loss for improved training. On the VisDrone-MOT dataset, MSTracker achieves state-of-the-art performance with 58.2% IDF1 and 41.0% MOTA, demonstrating significant effectiveness in challenging UAV scenarios.

源语言英语
主期刊名Proceedings of the 4th International Conference on Intelligent Computing and Next Generation Networks, ICNGN 2025
编辑Gyu Myoung Lee, Pavel Loskot, Qinmin Yang, Ruidan Su
出版商Institute of Electrical and Electronics Engineers Inc.
ISBN(电子版)9798331570705
DOI
出版状态已出版 - 2025
已对外发布
活动4th International Conference on Intelligent Computing and Next Generation Networks, ICNGN 2025 - Singapore, 新加坡
期限: 12 12月 202514 12月 2025

出版系列

姓名Proceedings of the 4th International Conference on Intelligent Computing and Next Generation Networks, ICNGN 2025

会议

会议4th International Conference on Intelligent Computing and Next Generation Networks, ICNGN 2025
国家/地区新加坡
Singapore
时期12/12/2514/12/25

指纹

探究 'MSTracker: A Multi-scale Spatio-temporal Framework for UAV Multi-Object Tracking' 的科研主题。它们共同构成独一无二的指纹。

引用此