Watch all the Transform 2020 sessions on-demand here.
Every year, Web Summit attracts celebrities from all over the world. This time, one of them was Ronaldinho, one of the greatest soccer football players of all time. Ronaldinho was promoting Teqball, a sort of mashup of football and table tennis played on a curved table. While I would love to spend this week’s column writing about how much fun my friends and I would have if we played Teqball all weekend, that’s not what stood out to me.
No. What struck me was that the creators of Teqball, football enthusiasts Gábor Borsányi and Viktor Huszár, concluded that they had to build an app called Sqiller to promote the new sport. And not just any app — one that uses machine learning and computer vision. Apparently, shooting videos isn’t enough in 2019. Having football stars Ronaldinho and Cafu as your ambassadors for a football-like sport isn’t enough. You need them, a mobile app, and some AI, plus a little gamification for good measure.
Here’s how it works. Professional football players create reference exercises by filming themselves. Players then try to match the “exercise” while someone films them. The game scores your performance — how close your touches are to the reference exercise — with a percentage score. You have to reach 90% to get to the next level. And you have limited lives — just like a typical mobile game.
June 5th: The AI Audit in NYC
Join us next week in NYC to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.
Teqball IT director László Bárdy told me that Sqiller uses an AI model running on the iPhone’s Core ML. He says their machine learning algorithm can detect a wide variety of balls and body parts. Analysis of the challenge happens in real time (players don’t have to upload their trick videos), with lag staying “within 0.5 seconds.” The app’s custom computer vision assesses ball touches and trajectories by modeling the path of the ball and factoring in the laws of physics. You don’t need a special ball, accessories, or to be in any specific environment. “We want it to be played anytime, anywhere,” said Bárdy.
The team taught the Sqiller engine about 100,000 different types of balls using images and videos of balls. “We also used synthetic data,” Huszar explained. “So we artificially created realistic images of the ball in certain life situations.” The app also has to work in different lighting conditions and make sure you’re actually performing the trick. It analyzes your skeleton to make sure you’re hitting the ball with, say your feet, not your hands. One level might only consist of headers, for example. You can’t cheat.
Sqiller launched in beta at Web Summit, but only for attendees and only on iOS. Huszar tells me that the official launch is set for January. He could not commit to an Android release, but he did spend plenty of time talking up his ball-tracking computer vision system.
ProBeat is a column in which Emil rants about whatever crosses him that week.
