Dedicated armchair coaches and micromanagers, you're in luck.
New video tracking can now show exactly who's where on the court or field. The computer system distinguishes between multiple players, making it easier for coaches, refs, and rabid fans to analyze what's happening.
"You could transpose this every time you have a team of people doing something together," said Pascal Fua, a computer science professor at the École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland and head of its Computer Vision Lab. He and a team of computer scientists presented the system this week at the International Conference on Computer Vision in Barcelona. "It is a noninvasive way of doing things."
The technology has a broad range of potential applications beyond sports. In a separate project, a surgical group is planning to use it to track the movements of everyone on the team in an effort to improve their workflow. At the conference in Barcelona, computer scientists discussed using the system to learn about ants' social patterns, research that has a bearing on effectively deploying robot swarms.
Using a computer to identify and follow individual players without help from Radio Frequency Identification chips or tags, which many sports federations don't allow, was an enormous challenge for the scientists. The field is packed with tracking systems, but the commercial actors don’t say what they are doing, Fua points out. The known systems usually rely on some sort of manual intervention, he added.
Unlike a human, a computer struggles to keep track of people walking in a crowd since sometimes people become virtually hidden by others. The same thing happens with players jockeying for position under a basket. To build their tracking system, the team used eight cameras trained on a basketball court. They also developed several unique algorithms to process the information from the cameras.
The algorithms can follow individual players just from the way they stand out from their surroundings. And they assign identities to individual players by reading jersey numbers the way traffic light systems capture the license plates of speeding drivers. Together, this allows the system to keep track of players even when they bunch together. Much like a video game interface, the tracker superimposes ID boxes over the players -- and coaches, and referees -- as they appear onscreen.
"In the basketball case, we have five players for each team, and typically three referees, and sometimes two coaches that are going inside and outside the court all the time. It's a huge mechanism," said Horesh Ben Shitrit (pronounced "Shih-treet"), a PhD student in Fua's lab who worked on the algorithms. The computer scientists are confident the tracking works with other sports including soccer, rugby, and American football.
This tracking also makes it possible to watch the way multiple shoppers move around a store and respond to factors such as product placement. The scientists think this information could be useful for marketing purposes.
It's too early for Fua to talk about commercializing the technology, though, he said. For now, his team is working on testing and finalizing the latest version of the system. Unlike the version presented in Barcelona, this one operates in real time.
Ramakant Nevatia is a computer science professor at the University of Southern California whose areas of expertise include pattern recognition and image processing. Sports players wearing similar jerseys and team colors who converge in small areas make camera tracking extremely difficult, he said.
"It is hard for me to say if the EPFL system will be able to track players accurately under all conditions, but the demonstrated results are very impressive," Nevatia said. "It is the most advanced system I know of at this time."
At the Georgia Institute of Technology, Aaron Bobick is professor and chair of the School of Interactive Computing. "Measuring the flow of people is just very difficult," he said. "This system can now handle that in a lot of different situations."
Accurate player tracking over time could potentially give coaches an advantage, Bobick said. "In soccer or football, if there was some connection between the result the particular plays had and movement of some players that wasn’t so obvious -- that's the kind of thing a machine could discover."