Reproduce “A Universal Protocol to Benchmark Camera Calibration for Sports” 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
Floriane Magera, Thomas Hoyoux, Olivier Barnich, Marc Van Droogenbroeck
Astradyne researchers from EVS Broadcast Equipment and the University of Liège introduced ProCC, a universal benchmarking protocol for sports camera calibration. This protocol, using semantic 3D point annotations and the JaC metric, revealed that camera models incorporating radial distortion provide significantly higher accuracy than traditional homography-based approaches, reducing 3D position estimation errors by over 2.5 meters compared to older methods.
Key insights
- Traditional homography-based ground truths and evaluation metrics are fundamentally inadequate for accurate camera calibration in sports, consistently underperforming more complex models.
- Richer camera models, such as a pinhole model extended with even one radial distortion coefficient, are crucial for achieving the necessary reprojection and 3D positioning accuracy in sports applications.
- Seemingly small 2D reprojection differences between camera models translate into massive and unacceptable 3D position estimation discrepancies (over 2.5 meters), failing professional accuracy standards for applications like offside detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.