Reproduce “PLayerTV: Advanced Player Tracking and Identification for Automatic Soccer Highlight Clips” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Håkon Maric Solberg, Mehdi Houshmand Sarkhoosh, Sushant Gautam, Saeed Shafiee Sabet, Pål Halvorsen, Cise Midoglu

PlayerTV presents an AI-driven framework for automatically generating player-specific highlight clips from soccer video footage. It integrates robust player tracking, team identification via color analysis, and kit number recognition, demonstrating high accuracy in team mapping (91.5% with CIELAB) but is currently limited by kit number OCR accuracy (30.6%).

Key insights

  • CIELAB color space clustering applied to refined regions of interest effectively differentiates teams based on kit colors, achieving over 90% accuracy.
  • Off-the-shelf OCR models are the primary bottleneck for accurate individual player identification via kit numbers in dynamic soccer videos, showing a peak accuracy of around 30%.
  • A modular pipeline design allows for future improvements in individual components, where enhanced OCR performance would proportionally boost the overall system's accuracy in player identification.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.