RefAI

video feed
app
techstack

Inspiration

This project is fueled by the same high energy as Marty Supreme: turning raw speed and split-second chaos into something precise, fair, and memorable. By layering real-time computer vision over game-state logic, we’ve built a system that gives every rally the honest, high-stakes clarity athletes and wagerers deserve. It’s more than just a tracker; it’s a trusted referee at the edge of the action; fast, confident, and always ready for the next point. Play and wager safely with Ref.AI.

What it does and how we built it

Ref AI pairs a unique computer vision model using colour masking to track the ball, along with a game-logic system that covers all possibilities of table tennis. We ensure that every point is properly counted and justified, giving players and wagerers peace of mind. To further build on our theme of fairness and security, we utilize Solana, along with our own script, which was implemented with the help of Google Antigravity. This employs a middleman to ensure all wagers are secure and have a guaranteed payout. By deploying our backend, consisting of the computer vision model and the game logic using Vultr and our frontend using Vercel, we create an atmosphere where everyone feels like they are in the game by providing a live feed. Controlling interactions between these two deployments ensures our application runs seamlessly and efficiently by utilizing two hosting platforms.

Tech Stack: https://imgur.com/a/h3ecHVP

Why it matters

As enthusiasm towards the sport of table tennis grows, disputes arise among players and fans alike. Our app offers a completely unbiased, fast-reacting, and accurate OpenCV model that is used to track the ball and make real-time decisions and referee calls. To add an extra bit of flair for the fans, we also implemented live and fully secure wagering where a middleman with no service fee is used, allowing users to truly step into the intensity of the game, without worrying about paying extra and insecure transactions.

Challenges we ran into

Like any project, Ref.AI came with its share of trials and tribulations. Our biggest challenge when creating Ref.AI was getting the camera to track the ball. We initially tried to use models online, but we found that they worked under very specific conditions and couldn’t handle the intensity of a real game. To overcome this hurdle, we decided to create our own model using colour masking, a method that isolates a specific colour for our model to detect. Another challenge was implementing Solana to its fullest extent so that it’s not just a risky and non-secure e-transfer for wagerers but instead uses a middleman to provide a sense of relief and security, demonstrating an understanding of and improving the implementation of modern technologies.

Accomplishments that we're proud of

This being our first time implementing many of the technologies used in our project, we’ve had some significant accomplishments. Our biggest accomplishment was implementing colour masking along with OpenCV to create a model that accurately tracks the ball. and without losing the ball. This not only tested our understanding of OpenCV but also forced us to think of solutions to the common problems most models face regarding tracking. Our second biggest achievement was creating a middleman (Escrow) for Solana by getting input from both sides of the betting party and interacting with the Solana API to create a secure transaction when the game concludes.

What we learned

As with any project, we had to learn many new things, the first of which was creativity in the form of finding ways to combine colour masking and OpenCV to create a model that tracks table tennis balls effectively. Moreover, we learned how to use the blockchain platform, Solana, to carry out transactions using a middleman created by us using Rust. Lastly, for this project, we learned how to deploy the frontend and backend of an application on different hosting software and have them interact to create a fully functional application.

What's next for RefAI

We hope that Ref.AI is just the beginning of what automated referees and secure wagering systems can evolve from. As we move forward, we aim to refine the model by using predictive tracking to account for edge cases where the ball is out of sight due to lighting, background interference, or the player blocking the ball. With this predictive tracking, players can also look back on their games and use our tool as an aid to their coaching. Overall, we strongly believe that with these additions and optimizations to the cloud hosting to accommodate multiple games and wagers, Ref.AI can become a staple in the table tennis community as a tool for players, enthusiasts, and coaches.