Davidsv/football_insight

Football Match Analysis

Model Description

This project uses YOLO v26 models for real-time football match analysis, including:

• Player detection and tracking
• Team classification by jersey color
• Referee detection (orange/yellow/pink jerseys)
• Formation line visualization
• Possession and pass statistics

Intended Use

• Sports analytics and coaching
• Match analysis and review
• Training video annotation
• Research in sports computer vision

Models Used

Model	Purpose	Size
yolo26m.pt	Player/ball detection	~50MB
yolo26m-pose.pt	Pose estimation (optional)	~50MB

Download models from Ultralytics.

How to Use

# Clone the repository
git clone https://github.com/YOUR_USERNAME/football-analysis.git
cd football-analysis

# Install dependencies
pip install -r requirements.txt

# Download YOLO models
# Models are automatically downloaded on first run

# Run analysis
python main.py input_video.mp4 output_video.mp4

Features

Team Classification

• Automatic team color detection using K-means clustering
• HSV-based referee detection (orange, yellow, pink jerseys)
• Temporal smoothing to avoid flickering

Formation Lines

• Detects horizontal player alignments (3+ players)
• Maximum 2 connections per player
• Visualizes defensive/midfield/attack lines

Statistics

• Real-time possession percentage
• Pass count per team
• Footer overlay with live stats

Configuration

Key parameters in config.py:

detection_confidence = 0.3  # Player detection threshold
ball_confidence = 0.2       # Ball detection threshold
device = "mps"              # Auto-detected: mps/cuda/cpu

Limitations

• Requires clear camera angle (broadcast view works best)
• Teams must have distinct jersey colors
• Possession stats are approximate (based on ball proximity)
• Performance varies with video quality

Performance

Device	FPS
Mac M4 (MPS)	15-20
NVIDIA RTX 3080	25-30
CPU (i7)	3-5

Training Data

This project uses pre-trained YOLO v26 models from Ultralytics, trained on:

• COCO dataset (person, sports ball classes)
• No additional fine-tuning on football-specific data

Ethical Considerations

• This tool is intended for sports analysis only
• Should not be used for surveillance purposes
• Respect privacy when analyzing non-public matches

Citation

@software{football_analysis,
  title={Football Match Analysis with YOLO v26},
  year={2026},
  url={https://github.com/David-dsv/football-analysis}
}

License

MIT License