arXiv 2602.004842026-02cs.CVcs.MM

GTATrack: 2025 SoccerTrack 挑战赛中采用深度EIoU和全局轨迹关联的胜出方案

GTATrack: Winner Solution to SoccerTrack 2025 with Deep-EIoU and Global Tracklet Association

Institute of Intelligent Systems\ of AI, National Yang Ming Chiao Tung University Tainan Taiwan；Department of Computer Science , National Yang Ming Chiao Tung University Hsinchu Taiwan；Technology National Central University Taoyuan Taiwan；Institute of Data Science\ Cheng Kung University Tainan Taiwan；Miin Wu School of Computing\ Cheng Kung University Tainan Taiwan；Institute of Intelligent Systems\ of AI, National Yang Ming Chiao Tung University；Department of Computer Science , National Yang Ming Chiao Tung University；Technology National Central University；Institute of Data Science\ Cheng Kung University；Miin Wu School of Computing\ Cheng Kung University

Rong-Lin Jian, Ming-Chi Luo, Chen-Wei Huang, Chia-Ming Lee, Yu-Fan Lin, Chih-Chung Hsu

备注：: Winner Solution of SoccerTrack in ACM Multimedia 2025 Workshop MMSports

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

GTATrack通过深度EIoU和全局轨迹关联技术，在2025 SoccerTrack挑战赛中以高HOTA得分和低假阳性率实现足球跟踪的最优性能。

摘要：

体育场景中的多目标跟踪（MOT）极具挑战性，由于球员运动不规则、外观统一以及频繁遮挡。这些困难进一步被静态鱼眼相机引入的几何失真和极端尺度变化所加剧。在本工作中，我们提出了GTATrack，一种分层跟踪框架，在2025 SoccerTrack挑战赛中获得第一名。GTATrack集成了两个核心组件：用于运动无关的在线关联的深度扩展IoU（Deep-EIoU）和用于轨迹级细化的全局轨迹关联（GTA）。这种两阶段设计使短时匹配和长时身份一致性都变得稳健。此外，使用伪标签策略来提升检测器对小而变形目标的召回率。局部关联与全局推理之间的协同作用有效解决了身份切换、遮挡和跟踪碎片化问题。我们的方法在HOTA得分上达到0.60的胜利成绩，并显著将假阳性减少到982，展示了在鱼眼相机基础上的足球跟踪中的最先进准确性。我们的代码可在https://github.com/ron941/GTATrack-STC2025上获取。

英文摘要：

Multi-object tracking (MOT) in sports is highly challenging due to irregular player motion, uniform appearances, and frequent occlusions. These difficulties are further exacerbated by the geometric distortion and extreme scale variation introduced by static fisheye cameras. In this work, we present GTATrack, a hierarchical tracking framework that win first place in the SoccerTrack Challenge 2025. GTATrack integrates two core components: Deep Expansion IoU (Deep-EIoU) for motion-agnostic online association and Global Tracklet Association (GTA) for trajectory-level refinement. This two-stage design enables both robust short-term matching and long-term identity consistency. Additionally, a pseudo-labeling strategy is used to boost detector recall on small and distorted targets. The synergy between local association and global reasoning effectively addresses identity switches, occlusions, and tracking fragmentation. Our method achieved a winning HOTA score of 0.60 and significantly reduced false positives to 982, demonstrating state-of-the-art accuracy in fisheye-based soccer tracking. Our code is available at https://github.com/ron941/GTATrack-STC2025.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。