Publication

Synergy between trajectories and human pose for soccer

Conference Article

Conference

IEEE International Conference on Image Processing (ICIP)

Edition

2026

Doc link

http://dx.doi.org/10.1109/ICIP61757.2026.11630385

File

Download the digital copy of the doc pdf document

Abstract

Predicting the future body poses of professional athletes is a critical task for sports analytics, yet the complexity of broadcast footage presents a significant challenge for traditional vision-based forecasting. In this paper, we propose a novel spatio-temporal framework that jointly forecasts player trajectories in pitch coordinates and 2D body poses in normalized image coordinates. Our approach leverages the bidirectional synergy between movement and articulation: global trajectories serve as a physical prior to pose predictions, while body poses encode movement information that improves trajectory forecasting. Our model uses a velocity-based contrastive loss to regularize the latent space, ensuring motion-consistent predictions across different intensity regimes. Experiments on a large-scale proprietary soccer dataset and on the public WorldPose benchmark demonstrate that our model significantly outperforms pose-only baselines.

Categories

computer vision.

Author keywords

Sports analytics, trajectory forecasting, human pose estimation.

Scientific reference

M. Peral, G. Capellera, L. Ferraz, A. Rubio, S. Grasshof, D. Witzner and A. Agudo. Synergy between trajectories and human pose for soccer , 2026 IEEE International Conference on Image Processing, 2026, Tampere, Finland.