Evaluating deep tracking models for player tracking in broadcast ice hockey video

TL;DR

This paper evaluates various deep learning-based tracking algorithms for player identification in broadcast ice hockey videos, addressing challenges like fast motion, occlusion, and camera movements, and providing performance analysis.

Contribution

It compares state-of-the-art tracking methods in hockey, highlighting their performance and failure modes, and bridges a gap in applying deep learning to this domain.

Findings

Deep learning methods improve tracking accuracy over traditional approaches

Analysis reveals common failure modes such as occlusion and rapid motion

Performance metrics for hockey player tracking are systematically reported

Abstract

Tracking and identifying players is an important problem in computer vision based ice hockey analytics. Player tracking is a challenging problem since the motion of players in hockey is fast-paced and non-linear. There is also significant player-player and player-board occlusion, camera panning and zooming in hockey broadcast video. Prior published research perform player tracking with the help of handcrafted features for player detection and re-identification. Although commercial solutions for hockey player tracking exist, to the best of our knowledge, no network architectures used, training data or performance metrics are publicly reported. There is currently no published work for hockey player tracking making use of the recent advancements in deep learning while also reporting the current accuracy metrics used in literature. Therefore, in this paper, we compare and contrast several state-of-the-art tracking algorithms and analyze their performance and failure modes in ice hockey.