Enhancing Soccer Game Analysis through Deep Learning-Powered Object Tracking and Statistical Evaluation | ||
| Port-Said Engineering Research Journal | ||
| Articles in Press, Accepted Manuscript, Available Online from 20 October 2025 | ||
| Document Type: Original Article | ||
| DOI: 10.21608/pserj.2025.410168.1430 | ||
| Authors | ||
| Mona Nashaat* ; Rabab F. Abdel-Kader | ||
| Port Said University | ||
| Abstract | ||
| Accurate object tracking in soccer game analysis is vital for deriving meaningful insights. Existing methods, however, face challenges due to occlusions, complex player interactions, and the dynamic nature of the game. This study presents FieldVisionAI, a novel deep-learning framework that integrates advanced tracking algorithms with statistical analysis to improve the precision and reliability of soccer match evaluations. Unlike traditional approaches relying on specialized camera equipment, FieldVisionAI works with standard TV footage. The framework comprises three key phases: object identification and classification, projection to a 2D field pitch, and detailed statistical analysis. By detecting and classifying players and the ball using color analysis, the framework handles occlusions effectively. The 3D tracking data is then transformed into a 2D field representation, facilitating visualization of player and ball movements. Finally, comprehensive insights, including distance covered, speed, player heatmaps, and expected threat values, are provided, highlighting the impact of specific actions on match outcomes. Experimental evaluations on real soccer footage demonstrate that FieldVisionAI significantly improves tracking accuracy and robustness, achieving up to 42.65% improvement in precision and 41.79% in recall. The proposed framework bridges the gap between raw tracking data and meaningful sports analytics, making it a valuable tool for analysts and coaches. To support transparency and reproducibility, the source code and a sample dataset have been made publicly available at https://doi.org/10.5281/zenodo.15163272. | ||
| Keywords | ||
| object tracking; object classification; ball and player detection; 2D projection; deep learning | ||
|
Statistics Article View: 1 |
||
