Professional sports are all about data and analysis. Sports science is a very busy subject and each professional team has to employ it for success. However, analysis might be quite tricky if everyone on the field looks almost exactly the same as in ice hockey. But you know what can help? Artificial Intelligence, as scientists at the University of Waterloo have discovered.
Waterloo is in Canada, which is famous for its involvement in ice hockey. And it is a beautiful game, involving precision, passion and incredible toughness. However, players are very hard to follow, because of identical uniforms and helmets. They are also changing all the time without stopping the game, which makes it even harder. Especially having in mind how fast-paced this game is. While fans can enjoy the game regardless of these issues, analysing the game is quite difficult and requires careful eyes.
And so scientists built a data set of more than 54,000 images from National Hockey League games. This was actually the largest data set of its kind. They then used this data to train their artificial intelligence (AI) algorithms to recognize the numbers on players’ sweaters. This is quite tricky, but also very important, because the number is basically the only major cue you have to identify a particular player in an ice hockey game. Sure, you can see the face a little bit too, but it is obscured by the helmet. However, the AI solution seemed to work well – researchers achieved an impressive level of accuracy.
Scientists managed to boost the accuracy by teaching their AI to consider two-digit numbers as two separate digits put together. While this was a simple idea, it helped crank up the accuracy of the algorithm to nearly 90 %. Kanav Vats, leader of the research project, said: “Using different representations to teach the same thing can improve performance. We combined a wholistic representation and a digit-wise representation with great results.”
Now researchers are developing AI, which will be able to track players in video, locate them on the ice and recognize what they are doing. The algorithm will be able to tell when the player is taking a shot or checking a player from the opposite team.
You may be wondering why AI is needed for this kind of thing in the first place. Well, analysis could be performed manually, but it takes a lot of work. Tracking each player in videos is very difficult, because they look nearly identical with their uniforms and helmets. But it is also needed for analysis of the tactics of the game as well as to illustrate some points during broadcast. It is also an interesting example of how useful AI solutions are in nearly every field of life.
Source: University of Waterloo