The demise of boring matches

No matter how fanatical, all football fans have seen themselves in a situation where they couldn’t wait for a match to finish, and the seemingly endless minutes stretched to infinity – not all games can be exciting for 90 minutes, after all.

However, British researchers are looking into a way to solve that problem by developing a computer algorithm capable of determining the interesting bits of a match automatically, without any human input. We spoke to David Chatting who works at BT’s Research and Venturing department and is behind the creation.

Computers understanding video

David works in the Broadband Applications Research Centre at BT: “Our job is to look at ways of increasing adoption of broadband and to look at interesting applications for them once they become available to people,” he explained. “Specifically, the group I’m in within that is the Media Interfaces Group. This has to do with finding ways to get computers to help us manage ever-growing amounts of content – we are only getting more videos and pictures, so how can we allow people to cope with that, find the thing that they’re after quickly and easily and in an enjoyable way?”

“One of the aspects of that is asking whether we can begin to make computers understand video: can we get them to start to index a video to find a particular point in there?” he said. “Our dream is to reach a point where we can say ‘find me the bit in this movie where that actor does a particular thing’ or the funny bit. A lot of that has to do with how does it make you feel: is it the exciting part, the sad bit or a humorous segment? It’s all about how the viewer reacts to that content.”

David and his colleagues have already succeeded in finding out what makes viewers of football matches tick, and are now able to extract what they call ‘affect’, the mood of a particular moment in a game. “We take the video of the match and we look at three different aspects of that video to try and determine the moments when excitement peaked, and those tend to be – in the case of football – things like goals and other incidents. Firstly, we look at the sound of the crowd as they are always reacting to things; if you had the TV on in another room and there’s been a match on, you’re always aware of what’s happening even if you’re not watching it. We look for that kind of energy in the higher frequencies of the audio.”

Camera angles and on-screen movement

He continued: “Secondly, we look at how often the camera changes as we can detect when the shot changes. You tend to have that sort of sweeping, panning shots during the majority of the match, and then something interesting happens and they’ll cut to a different camera. Maybe it’s a close-up of a player’s face and then you cut to the manager’s face or the goalkeeper’s face – you have a much higher rate of cuts around something that’s interesting. We look at the frequency of those cuts, so that’s our second feature.”

The third feature, according to David, is how much motion is happening in the scene at any particular moment. If the picture is showing someone standing still in the middle of the pitch, it is possible to infer nothing is happening, but if there is a lot of activity something is bound to be going on.

“We combine all those features together to get an overall score of interest, of predicted excitement,” David added. “We have a graph of time going from zero to 90 minutes, and for each of those moments we can plot what we predicted to be the excitement of that moment. What you get is a graph where the peaks will be around interesting moments, such as goals and major incidents.”

Looking into the future

The immediate application of this, David said, is to create a highlights programme showing only the most exciting moments of a match, which can be condensed down to one or two minutes’ worth of the most interesting activity as predicted by their algorithm. “As it stands, we’ve done it with quite a few matches now and it works – we reliably get out those goals and interesting moments,” he explained. However, the algorithm’s success doesn’t mean their work is done: “There are research questions to consider; a highlights programme isn’t just a collection of moments that you cut and put together arbitrarily. There are editing decisions that are necessary to tell the story [of the game]. In a way these are quite crude highlights, but they do show you those key moments.”

And what does the future hold for this project? “We’re currently thinking of other things in which to use that, such as offering a ‘smart index’ to the match so that you are able to fast forward to the next interesting moment, without any radical cuts,” answered David. “That can be applied to matches that happen today, but also to archive material. Ultimately, we would like to be able to apply that to more complex content – the nice thing about football matches is that the exciting moments are pretty well-defined. We can all agree that goals are interesting, whereas if we start talking about films it gets more subjective.”

"Ultimately, we would like to be able to apply that to more complex content – the nice thing about football matches is that the exciting moments are pretty well-defined. We can all agree that goals are interesting, whereas if we start talking about films it gets more subjective."

David Chatting

Despite the size of the challenge, David is already thinking ahead: “What I’m working on now is trying to understand what a real person feels when they are watching [films]. We’ve been using so-called physiological readings, things like the way people’s skins sweat – not unlike a lie detector. We measure the resistance of people’s skins as they watch the content and you see that when they see something interesting and their attention is peaked, you have a peak in the graph as well. I’m looking to see if you can correlate those peaks with our predictions to be able to assess whether we’re doing a good job or whether we can improve our algorithm in some way, using this as grounding.”

You’ve read it. Now review it.

Back to top

Back to top