Hawkeye

[ZacNaloen]

Prediction was possibly a poor choice of word but I'm also struggling to think of a better one , it's measuring the balls flight and determining what it thinks it is doing. If it thinks it wholly crossed the line it beeps. It doesn't even need to see it happen.

For example the ball goes towards goal, someone gets a block on the ball, but the camera loses sight of the ball for the second that it is on that line as the player blocked the view. Doesn't matter, it can calculate what probably happened based on what the ball did next.

That's the predictive nature of it. This prediction is less likely to be accurate than if it had a full view. The accuracy of that prediction is what has been everyone's worry with it, but they've obviously demonstrated sufficient accuracy or they wouldn't have got the license. Or they just added more cameras to reduce the likely hood.

[Cobi]

Prediction was possibly a poor choice of word but I'm also struggling to think of a better one , it's measuring the balls flight and determining what it thinks it is doing. If it thinks it wholly crossed the line it beeps. It doesn't even need to see it happen.

For example the ball goes towards goal, someone gets a block on the ball, but the camera loses sight of the ball for the second that it is on that line as the player blocked the view. Doesn't matter, it can calculate what probably happened based on what the ball did next.

But what happens when it loses sight of the ball and the ball is handled, knocked on, interfered with etc? It still 'guesses' but wont take into account the new trajectory of the ball.

[Alexander Litvinenko]

Indeed, it's not doing predictions, it's doing actuals, and there's a whole lot more data to process and things to get in the way or confuse it with arms, legs, heads and other body parts also potentially being over the line or in the way at the same time.

A "prediction" won;'t work because, as Cobi says, the ball might be deflected in the last millisecond before crossing the line.

[ZacNaloen]

Prediction was possibly a poor choice of word but I'm also struggling to think of a better one , it's measuring the balls flight and determining what it thinks it is doing. If it thinks it wholly crossed the line it beeps. It doesn't even need to see it happen.

For example the ball goes towards goal, someone gets a block on the ball, but the camera loses sight of the ball for the second that it is on that line as the player blocked the view. Doesn't matter, it can calculate what probably happened based on what the ball did next.

But what happens when it loses sight of the ball and the ball is handled, knocked on, interfered with etc? It still 'guesses' but wont take into account the new trajectory of the ball.

The system has to be able to cope with not being able to see the ball for a second or so It has to. otherwise it's not fit for purpose. To do this it will constantly be running calculations to determine information like this:

Which pixels are most likely to be the ball. Where the line is. Are those pixels moving towards the line. Will it cross the line. When do I expect it to cross the line. How fast is it moving.What it's trajectory. Did it cross the line. Where is it now. Could it have crossed line in the time it took for it travel from it's previously known position to it's current position at the current velocity and trajectory.

Indeed, it's not doing predictions, it's doing actuals, and there's a whole lot more data to process and things to get in the way or confuse it with arms, legs, heads and other body parts also potentially being over the line or in the way at the same time.

A "prediction" won;'t work because, as Cobi says, the ball might be deflected in the last millisecond before crossing the line.

It cannot work without using predictions. It's impossible. By the very nature of any camera footage being analysed there will be an element of prediction in the algorythms being used.To say it's using "actuals" is simplistic. A computer has no idea what reality is, it cannot use actuals. It has no idea what this is. It can make a prediction, using complicated algorythsm. Using maths I definitely don't understand, to do very high accuracy guesses. Nothing more.

[Alexander Litvinenko]

That's why, technologically, it's far more complex than the tennis and cricket systems.

It can work without predictions, but only by using a system that has so many sensors that the odds of more than 2 of them being obstructed are greatly reduced (but not completely eliminated, of course).

Plus, another complication that you don't get in the tennis and cricket systems is that it needs to work in conditions of heavy rain and potentially snow, at a variety of temperatures. And it needs to take into account factors like the fact that the crossbar (on which sensors might be mounted) might be moving/vibrating if a shot hits it and the ball then goes down on/over the line.

As I say, complex technology, but Hawkeye reckon they've got it nailed.

Utilising the latest technology of high frame rate cameras and cutting edge vision processing techniques, combined with over 12 years of experience in sports officiating and broadcast enhancement, Hawk-Eye are well placed to provide a Goal Line Technology system for football matches.

Hawk-Eye has now signed a licence agreement with FIFA, giving Hawk-Eye full authorisation to install the system worldwide. In order to become a FIFA licensee Hawk-Eye met a range of system requirements and underwent a series of tests in match-like conditions and has been tested at St Mary’s Stadium, home of Southampton Football Club.

The current system boasts the following features:

Accurate – betters the FIFA requirement of +/-3cm.
Reliable – the system is fully automated and benefits from our unique and extensive experience of delivering robust officiating systems for elite Tennis and Cricket matches. Only a small percentage of the ball needs to be visible in the cameras for the system to accurately locate it.
Non-Invasive Hawk-Eye works without the need to adapt the ball, goal posts or playing surface.
Ball agnostic - Hawk-Eye works with any ball manufacturer.
Fast - An outcome can be relayed to the referee within 1 second.
Compliant - Fully complies with FIFA specifications.
Definitive - Hawk-Eye can provide a TV replay to categorically prove the decision is correct.
Robust - No single points of failure.
Occlusion - Independently tested to work in cases where the ball is more obscured than has ever occurred in a real match goal scenario.
Secure - Hawk-Eye have exclusively developed a referees watch in partnership with “Adeunis”, the market leader in stadium communication systems. The signal to the watch is encrypted, unaffected by interference and works in all countries.

Hawk-Eye is confident that the days of post-match retributions will soon be a thing of the past. The closest goal-line calls will be resolved in a matter of seconds as Hawk-Eye’s experience and accuracy on the tennis court will take football officiating to a new level.

From the line "Only a small percentage of the ball needs to be visible in the cameras for the system to accurately locate it" I assume it has recognition systems that know what a ball looks like (size and shape) and so extrapolate the rest of it into a 3D virtual image.

[ZacNaloen]

That IS a prediction!

[Alexander Litvinenko]

I think we're into semantics now.

If it *knows* where part of the ball is, even just a small part of it, then calculating where the rest of it isn't (to me) a prediction.

But if it doesn't know where any of it it right now, but it knows where it was and where it was going, then that is a prediction - not allowed with this system.

[ZacNaloen]

We are arguing semantics.


Because it is a prediction.

[Alexander Litvinenko]

But if it is a prediction it's based on minimal variables - it's predicting the position based on the location of the rest of the ball based on the knowledge of where part of it is, and what size and shape the ball is. They're very, very, very unlikely to be wrong.

But if the prediction is based on previous known location of the ball and it's movement then there are far more variables, so that prediction will have much less certainty behind it.