Barcelona

Football broadcast-to-tracking pipeline for extracting player and ball coordinates from match clips.

What this repo does

• Detects players and ball from broadcast video.
• Tracks objects across frames.
• Projects coordinates onto a standard 105x68 pitch.
• Produces annotated video and structured outputs for analysis.
• Exports summary_stats.json with per-player distance, average speed, team totals, and ball travel distance.

Quick start

1. Install uv.
2. Install dependencies:

   uv sync

3. Run inference:

   uv run main.py --video_path input_video.mp4

4. Check output under output/<video_name>/.

Extra options

• Tune movement filtering for stats:

  uv run main.py --video_path input_video.mp4 --max_step_meters 2.5

• Skip stats export:

  uv run main.py --video_path input_video.mp4 --disable_summary_stats

Useful paths

• Code: eagle/
• Examples: examples/
• Documentation: docs/
• Outputs: output/

Notes

• GPU is strongly recommended for faster inference.
• Input clips with stable camera angles generally produce better results.

Reference: https://github.com/nreHieW/Eagle