2023 has been a defining year for artificial intelligence.
Tennessee head coach Josh Heupel and LSU head coach Brian Kelly embraced AI at media days, discussing intriguing opportunities to implement the technology within their football programs.
Professor Dr. D.J. Lee has already kicked off studying AI at BYU. He is part of BYU’s Department of Electrical and Computer Engineering.
Lee discussed the early process of studying AI in relation to football with Vols Wire.
“We started this about two, two-and-a-half years ago,” Lee said. “My research has always been AI for visual information and processing, so I thought about applying it to football. We went and talked to the BYU football team and tried to learn what they are doing and to see what we could do to help them. They are still using software to do manual innovative of play formations based on videos.”
SEC media days: Josh Heupel discusses AI becoming a resource in college football
Lee’s early analysis of studying video provided to him is that camera angles are the most important aspect for teams to gain a competitive edge with AI.
Lee mentioned that sensors are not required, only an overhead camera of at least 50-70 degrees. The precise camera angles are a critical part of continuing a process of implementing AI into football programs.
“We started by looking at video they provided,” Lee said. “The next step will be to find videos recorded from higher up. Ideally, it would be having the camera placed up high to see the whole football field and does not move. That way we can use our algorithm to locate a player, determine the positions and the overall formation of the offense.
“We can then track the players, each player, and know how they move. We can also have them cut out the video for the part that they are interested in, so you don’t have to look at the videos during substitutions, between plays, or timeouts, all of that can be removed or find a time stamp to look at short video intervals you are interested in. There are a few things we can do to help football teams if a camera is mounted higher up.”
Lee also has relied on a Madden NFL video game as he continues to study AI to help teams.
“We bought the Madden 2020 video game,” he said. “You can select the formation and we captured the images and the videos from the game. We collected hundreds of videos and images to test it out. The result was very good, but we also found some challenges. The biggest challenge is the camera and occlusion problems.
“The quarterback maybe blocking the center, the center maybe blocking the defense. It is not enough to see every player, so we have to do a lot of work to try to see where all 22 players are. That was the biggest challenge using the Madden video game for collecting data to train our algorithm. The other challenge using it, once you hit play and the ball is snapped, then the camera moves to zoom into the ball carrier most of the time. You lose track of other players and how they move, so it’s not ideal. We managed to get some very promising results that we could recognize offensive formations.”
Lee’s next step is to have cameras recording from a higher position to collect more data, including tracking movement of players.
“You can track the players and see how they move,” Lee said. “That way it will help coaches to better game plan and is also a good training tool for the players to see how the opponent’s defense reacts during certain situations. With more video, I can make a dynamic game chart. I think that would be very helpful to prepare for players to improve their performance.
“Right now, we can accurately recognize formations from an image, but videos with cameras from high up can get to a dynamic play chart fairly quick.”
SEC media days: Brian Kelly discusses intriguing opportunities with AI in college football
[affiliatewidget_smgtolocal]