Scholarlink

<p>Multi-object tracking (MOT) has received extensive research and attention in the fields of graphic image processing and multimedia vision in recent years due to its great application value in business. The task of MOT tends to localize multiple objects in a video and trace their trajectories. In complex imaging conditions, where the degradation is diverse and unknown, it is non-trivial to identify and track the objects accurately under adverse weather scenarios. To address this problem, this work harmonizes the merits of the blind super-resolution (BSR) technique to the MOT domain to improve tracking accuracy. Specifically, we use the real-world blind super-resolution technique to improve the recognizability of the focused objectives with better details and clarity in graphic images, which provides more clues for the tracking pipeline. In addition, to achieve more robust tracking, this work further designs a clustering learning module, which reduces the complex computation and achieves more accurate association to improve the tracking performance. Extensive experiments have been conducted on publicly available datasets (MOT16 and MOT20) to verify the performance of our proposed method, which can also effectively meet the practical requirements of graphics and image visualization applications. The code is available at&#xa0;<a href="https://github.com/szh91/SRCA">https://github.com/SRCA.</a></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Online visual multi-object tracking via blind super-resolution and clustering learning

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Zhihong Sun,
  • Zheng Lv,
  • Mingfu Xiong,
  • Wei Hu,
  • Guoheng Wei,
  • Jiuxing Zhou,
  • Tao He,
  • Qingrui Meng

摘要

Multi-object tracking (MOT) has received extensive research and attention in the fields of graphic image processing and multimedia vision in recent years due to its great application value in business. The task of MOT tends to localize multiple objects in a video and trace their trajectories. In complex imaging conditions, where the degradation is diverse and unknown, it is non-trivial to identify and track the objects accurately under adverse weather scenarios. To address this problem, this work harmonizes the merits of the blind super-resolution (BSR) technique to the MOT domain to improve tracking accuracy. Specifically, we use the real-world blind super-resolution technique to improve the recognizability of the focused objectives with better details and clarity in graphic images, which provides more clues for the tracking pipeline. In addition, to achieve more robust tracking, this work further designs a clustering learning module, which reduces the complex computation and achieves more accurate association to improve the tracking performance. Extensive experiments have been conducted on publicly available datasets (MOT16 and MOT20) to verify the performance of our proposed method, which can also effectively meet the practical requirements of graphics and image visualization applications. The code is available at https://github.com/SRCA.