A player tracker for Roller hockey games.

⚠️ only tested with fixed high-resolution camera.

Requirements:

• python3
• ffmpeg for clip and frames extraction
• youtube-dl to download the video
• matplotlib to display the trajectories
• opencv-contrib-python for some image processing tools
• torch, torchvision for player detection and other neural network things
• sklearn for other machine learning tools
• numpy, tqdm

First you need to create a project:

python -m rhtrack.project https://www.youtube.com/watch\?v\=SOME_ID PROJECT_NAME

Then you can process a clip, say a 8 seconds clip starting at 32:22:

python -m rhtrack.clip --start=00:32:22 --duration=8 --num-workers=5 \
    --project=PROJECT_NAME clip-1

This will create trajectories chunks to label with player names.

Open the file PROJECT_NAME/clip-1/trajectories/players.txt :

0: player_name
1: player_name
...

Each index corresponds to an image in PROJECT_NAME/clip-1/trajectories/.

Open the image to identify the player and replace player_name on the corresponding line with the name of the player.

Leave empty (i.e. just remove player_name) if you don't want this trajectory in the final video.

Once you're done, run:

python -m rhtrack.render --project=PROJECT_NAME clip-1

which will create the video: PROJECT_NAME/clip-1/output.mp4.