Abstract
We propose a computationally efficient method for real-time three-dimensionalfootball trajectory reconstruction from a single broadcast camera. In contrastto previous work, our approach introduces a multi-mode state model with $W$discrete modes to significantly accelerate optimization while preservingcentimeter-level accuracy -- even in cases of severe occlusion, motion blur,and complex backgrounds. The system operates on standard CPUs and achieves lowlatency suitable for live broadcast settings. Extensive evaluation on aproprietary dataset of 6K-resolution Russian Premier League matchesdemonstrates performance comparable to multi-camera systems, without the needfor specialized or costly infrastructure. This work provides a practical methodfor accessible and accurate 3D ball tracking in professional footballenvironments.