Resources
Highlights
Summary
In this group project, we developed a real-time vision-based football analysis system using a fine-tuned YOLOv11 object detection model, capable of identifying players, goalkeepers, referees, and the ball directly from high-resolution match footage.
The system achieves a mean Average Precision (mAP@50) of 0.827 across the main object classes and runs at 30 FPS on a consumer-grade RTX 2060 Super GPU, demonstrating the feasibility of low-cost, real-time sports analytics.
Two core components were built:
Object detector: A YOLOv11 model was fine-tuned using a manually cleaned and augmented dataset of 400 labeled football frames. Using 5-fold cross-validation, the ‘m’ variant was selected for its balanced accuracy and efficiency. Despite excellent performance on most classes, ball detection remained a challenge due to its small size and visual ambiguity.
Team classificator: An unsupervised pipeline was designed to assign detected players to teams using K-means clustering on torso color features. Without prior knowledge of jersey styles, the system achieved visually coherent team groupings, generalizable across matches.
The project highlights the potential computer vision-driven sports intelligence in accessible environments such as local clubs or amateur leagues, and lays the groundwork for more advanced features like event detection and spatial reasoning in future versions.
Pictures
Mean ± standard deviation of AP@50 per class, for each candidate model
Normalized confusion matrix for the selected model ’m’, indicating class-wise accuracy and common misclassifications
Precision–Recall curves for the selected model ’m’. A confidence threshold near 0.2 offers a good trade-off between precision and recall
Sample prediction made by the selected model ’m'
See it in action