Skip to main navigation menu Skip to main content Skip to site footer

Vol. 11 No. 1 (2025)
Special Issue: Proceedings of CVIS 2025

Articles

Video-Based Player Re-Identification in Ice Hockey via Non-Contextual Implicit Features

https://doi.org/10.15353/jcvis.v11i1.10001

Published 2026-04-01

  • Evan Iaboni
    • ,
  • Amir Nazemi
    • ,
  • Yuhao Chen
    • ,
  • David Clausi

Abstract

Player re-identification (ReID) in ice hockey is difficult due to similar uniforms, motion blur, occlusions, and obscured jersey numbers. We introduce a video-based method that focuses on extracting implicit features by combining OSNet spatial features with a lightweight temporal transformer. Using an ice hockey dataset, our approach benefits from simple data augmentations and outperforms state-of-the-art video ReID models in a zero shot setting by +4.7% mAP and +4.3% rank-1 accuracy. Analysis of the learned embeddings shows that the implicit and non-contextual features learned by the model are efficient enough to capture explicit attributes such as team, handedness, and jersey number.
PDF