I’m want to share with you guys a project I've been doing at Uni with one of my professor and that isFutbol-ML our that brings AI to football analytics. Here’s what we’ve tackled so far and where we’re headed next:
What We’ve Built (Computer Vision Stage) - The pipeline works by :
Raw Footage Ingestion • We start with game video.
Player Detection & Tracking • Our CV model spots every player on the field, drawing real-time bounding boxes and tracking their movement patterns across plays.
Ball Detection & Trajectory • We then isolate the football itself, capturing every pass, snap, and kick as clean, continuous trajectories.
Homographic Mapping • Finally, we transform the broadcast view into a bird’s-eye projection: mapping both players and the ball onto a clean field blueprint for tactical analysis.
What’s Next? Reinforcement Learning!
While CV gives us the “what happened”, the next step is “what should happen”. We’re gearing up to integrate Reinforcement Learning using Google’s new Tactic AI RL Environment. Our goals:
Automated Play Generation: Train agents that learn play-calling strategies against realistic defensive schemes.
Decision Support: Suggest optimal play calls based on field position, down & distance, and opponent tendencies.
Adaptive Tactics: Develop agents that evolve their approach over a season, simulating how real teams adjust to film study and injuries.
By leveraging Google’s Tactic AI toolkit, we’ll build on our vision pipeline to create a full closed-loop system:
We’re just getting started, and the community’s energy will drive this forward. Let us know what features you’d love to see next, or how you’d use Futbol-ML in your own projects!
We would like some feedback and opinion from the community as we are working on this project for 2 months already. The project started as a way for us students to learn signal processing in AI on a deeper level.
Hello!
I’m want to share with you guys a project I've been doing at Uni with one of my professor and that isFutbol-ML our that brings AI to football analytics. Here’s what we’ve tackled so far and where we’re headed next:
What We’ve Built (Computer Vision Stage) - The pipeline works by : Raw Footage Ingestion • We start with game video.
Player Detection & Tracking • Our CV model spots every player on the field, drawing real-time bounding boxes and tracking their movement patterns across plays.
Ball Detection & Trajectory • We then isolate the football itself, capturing every pass, snap, and kick as clean, continuous trajectories.
Homographic Mapping • Finally, we transform the broadcast view into a bird’s-eye projection: mapping both players and the ball onto a clean field blueprint for tactical analysis.
What’s Next? Reinforcement Learning! While CV gives us the “what happened”, the next step is “what should happen”. We’re gearing up to integrate Reinforcement Learning using Google’s new Tactic AI RL Environment. Our goals:
Automated Play Generation: Train agents that learn play-calling strategies against realistic defensive schemes.
Decision Support: Suggest optimal play calls based on field position, down & distance, and opponent tendencies.
Adaptive Tactics: Develop agents that evolve their approach over a season, simulating how real teams adjust to film study and injuries.
By leveraging Google’s Tactic AI toolkit, we’ll build on our vision pipeline to create a full closed-loop system:
We’re just getting started, and the community’s energy will drive this forward. Let us know what features you’d love to see next, or how you’d use Futbol-ML in your own projects!
We would like some feedback and opinion from the community as we are working on this project for 2 months already. The project started as a way for us students to learn signal processing in AI on a deeper level.