The Sports Measurement Problem

Luckily for everyone, I'm a sports engineer and not an England International. But I'd love to measure exactly what's going on when our International footballer takes a penalty under those extreme tournament conditions.

Here in the biomechanics lab, we can take accurate measurements of movement. The cameras and lights are on, I'm wearing my black suit, I've got passive markers all over me. I can sort of visualise the goal in the distance, but the problem is it's just not that realistic. It's like that problem in quantum mechanics; the act of measuring something changes it.

What we really want is a cheap system that we can use out in the field that measures the player without changing what he does. One way of doing that is to take a video of a sporting performance. And here we have a grey scale video of a single tennis shot during a game.

We've tracked the ball, and we can look at the brightness of the pixels surrounding the ball with time. And that gives us information on speed, spin, and angle of that shot, all with a single camera.

Now that's great, but it takes a lot of number crunching, and we can only get the answer after the performance. And as we've seen with our previous film on diving, we really want instantaneous feedback. And what about bigger and more complex sports like football?

Like any good engineer, we've been looking at the tools for a solution, and we've been exploring a piece of technology from a very different source of game, the Xbox Kinect.

John, you've been working on the Kinect sensor for 18 months, and this is the first time we've had the opportunity to plug it into an Xbox and play a game with it. Tell me how it works.

Obviously, for us to be able to play this game, what the Kinect's doing is working out where we are in three dimensions. And the way that it does that is an infrared projector on the sensor sends out a speckled pattern of infrared dots into the scene, and compares the image that it gets from its infrared camera to an image that's stored on the device. And the differences between those two images tell the Kinect not only where we are in the image plane, the horizontal distances, but where we are in terms of depth away and towards the Kinect.

In the lab here we've got a really expensive bit of kit that does exactly the same thing, but costs 1,000 times as much. So please tell me that we haven't wasted our money.

Well, the accuracy of the Kinect looks pretty good. We might be interested in the location of a badminton player on a court, for example. So we can locate the player to around about 5 to 10 centimetres of error, which, in the context of the size of the badminton court, is pretty good.

But for finer measurements, the sort of measurements that we might take in here with the system, we might be interested in measuring joint angles or the positions of body segments, we're not getting the levels of accuracy that we would see in the lab.

But the benefit of the Kinect is that it's cheap. It's essentially a commodity item, and we can take the Kinect out and take measurements in the field, on a player's badminton court, and essentially where they're performing.

How long before we have one of these out on a football pitch?

It's difficult to say, because the drivers for this technology have obviously been the sort of thing that normally happens in a living room, over quite small areas. So it's difficult to know what the drivers will be for the development of the technology to work over larger areas.

Whoa. Who's winning?

I have no idea, Steve.

We can take accurate measurements within the laboratory, but we're really constrained by the environment. Where we'd really like to be is out here in the field taking measurements of the reality of our athletes.

Now that's a tremendously challenging thing to do, but with image processing techniques, with high definition cameras, with depth cameras, combined with mobile technologies, I think we're on the verge of being able to do that. And for a sports engineer, that's tremendously exciting.