How accurate is VAR in Norway?

Pitch measurement

The courses are measured by Mitta, an international surveying company. If a club changes the pitch surface, the pitch must be measured again. Markers are set out and there is every reason to believe that this measurement is accurate to the millimeter level, in the same way as when measuring land for building structures, bridges, etc.

Potential deviation when measuring the pitch surface: Millimeter precision

Software calibration

The TV cameras sweep over the pitch ahead of the match to calibrate the pitch markings. Operators can adjust the points and the lines so that the software EVS at the bottom can draw the pitch’s dimensions in 3D over several camera angles. It is this software that, in turn, tells you how big an offside is in centimeters and millimeters. The measurement between the software’s own calibration (points and lines in the pitch surface) is considered to be accurate.

Potential deviation when calibrating the software: Millimeter precision

Camera synchronization

A person throws a ball on the ground visible to all cameras, so that you also synchronize slow-motion cameras that may be a frame or two behind and do not produce the same time code with several frames per seconds. This is the same as a clapper in a movie. All the cameras now have a reference point, in the same way as the moment a ball is played.

But at the same time, there is a limit to how many images per second the cameras capture. Apart from the slow-motion cameras, most are 50i, which merge to 25p in the VAR room. This also applies to the offside cameras at every box. This means that for every second of video, you have 25 images to look at. This again means that there are 40 milliseconds between each image.

Deviation between 2 frames 25p: 40 milliseconds

Video resolution

The TV images are 1920 pixels wide and 1080 pixels high. This means that in some angles you get several centimeters per small pixel on the screen. For example, 1 cm, i.e. 10 times «millimeter precision» in the image below, corresponds to 0.37 pixels. No screen can show half pixels. 1 pixel can only manage to display 4 cm in this case. But there is also motion blur, and the fact that an object is rarely in the middle between two pixels. The edge of the object will thus bleed outwards also between pixels.

In extreme cases with high contrast, you may also experience some chromatic aberration, particularly at the edges of the image, where the optics in the cameras are unable to concentrate all the light waves correctly.

Potential deviation in a wide shot in full HD: 40 millimeters given a completely sharp image (not TV or VAR)

Motion blur and movement between frames

Each image needs to be exposed, meaning that the sensor receives light for a certain time. Often this is about half the time between two images, so that it is not perceived as choppy for us to look at. This means that objects that are stationary relative to the camera are sharp, while moving objects are blurred.

When filming under artificial light on alternating current in PAL areas, the shutter speed should match the frequency of the mains (50Hz) to avoid flickering. So we usually film TV with 25 progressive images per second (the entire image is captured at the same time) with a shutter speed of 1/50 second. Matches are usually filmed in 50 interlaced, where the camera captures the image twice as often but only draws every second line each time. This is an «old» way of sending a picture stream in linear TV with half the bandwidth. When this is combined to 25p, each image becomes a combination of two «half» images. If you have seen old TV clips uploaded to YouTube, you can often see horizontal lines, especially in motion. This is interlaced material that has not undergone de-interlacing. The VAR room has stated that they are watching 25p on their screens. Both 25p and 50i usually use a shutter speed of 1/50s.

Motion blur means that by being a bit unlucky with a movement, both as a player and with the camera, the movement can «bleed out» in the image for 20 milliseconds. It sounds little, but measured in millimeters it is significant. A player in a full sprint runs at 8-9 meters per second. That is, during the time a picture is taken, he moves 16-18 cm (8000 mm * 0.02 sec).

Body parts also sometimes move faster than the body itself, for example the ankle will be thrown forward in a full sprint. This means that a foot at full speed forward can have a speed of up to 21.2 ± 6.8 m/s. This applies to the moment of the shot, so as a precaution we can set an assumed running speed somewhat lower when the player is just running. Let’s say the very lowest speed at a shot that is 14.4 m/s. At this speed, a running foot will move 28.8 cm in one frame and 57.6 cm between 2 frames. (14,400mm * 0.02sec)

Other parts of the legs are illustrated below.

Potential deviation running player: 160-180 millimeter in one frame

Potential deviation running player: 320-340 millimeter between two frames

Potential deviation running foot: ± 288 millimeter in one frame

Potential deviation running foot: ± 576 millimeter between two frames

Another and much faster object is the ball. The fastest shot I’ve found had a speed of 210.9 km/h, which is quite extraordinary. It is more common for shots and hard passes to be 70-100 km/h. I include shots here on purpose because offside is often also decided on deflections. The same survey shows that the average shot speed is 27.9 ± 1.3 m/s.

In other words, the ball moves an average of 55.8 cm from the foot and in the direction of the shot in a single image. This makes the ball oblong and blurry.

A ball that is shot or played hard can thus move 111.6 cm between 2 images. This says something about how difficult it really is to find the «moment of the pass«, the point where you have to decide offside or not, when working with the video material in the VAR room.

You can easily find shots that move a much greater distance between 2 images. If you retrieve clips from the Media Bank and scroll between 2 frames, as below, a completely normal shot can move closer to 2 metres.

You can see that the ball is slightly less blurry since the camera pans in the same direction.

Potensial deviation for a ball at speed: ± 1116 millimeter between two frames (moment of passing or playing the ball)

Real-world examples

This was a lot of numbers and theory, so let’s look at a real-world example. The factors mentioned above can work both with and against each other, so the variation in accuracy is quite large and you have to look at the whole picture to assess the margin of error.

In the case of the match between KFUM and Glimt, 2 different images were shown of the offside situation where the winning goal was disallowed by VAR. TV2 showed one frame during the match, which is the passing moment – or as close as you get. We know this because the ball comes back from the goal as a deflection, stops on the foot of a Glimt player and is played back to Kasper Høgh. Then we have 3 frames. Blurry ball on its way to foot, round ball on foot and blurry ball that has left foot.

Sorry for the Norwegian in this illustration. It reads:

Top:

The ball is round because the frame captures the moment the ball is played. The ball comes back from the goal, and is hit by the player on the volley. This frame was used in the broadcast.

Bottom:

The ball is blurry because the frame captures the ball after it has been played. This is the image published by the VAR room.

Here I would like to argue that the VAR room chooses the wrong frame to work with. This gives the attacking player a disadvantage, because he is at full speed forward while the defender is stationary in the sense that he has his foot on the ground. Høgh thus has 40 milliseconds of extra movement where he can potentially end up offside.

Another variable we have discussed here is the speed of body parts. The player is heading towards the goal at such a speed that he loses his footing and slips when he is about to turn. In this movement, he lifts his left leg.

In the example below, we see that Høgh is placed many centimeters further towards offside simply because the VAR room chooses a frame from after the ball has been played. When I was in the VAR room, we drew on this frame first an offside of 35mm, then 39mm, before VAR chief Svein Oddvar Moen confirmed 50mm – one month after he stated it was 355mm and not even a close call.

Based on how long a foot in size 45 is (28.6cm), we can assume that Høgh was actually more than 5cm onside at the moment of the pass.

Sorry for the Norwegian in this illustration too. It reads:

Top:

Høghs foot is further behind when the ball is played. This is the image from the broadcast.

Bottom:

Høghs foot is closer to the goal (and closer to offside). This is the image published by the VAR room.

Points and height above the ground is set manually

The Replay Operator is the one who has his hands on the levers to set points and lines, but it is the VAR referees who approve where these are set. As mentioned above, there will be variations of several centimeters in the video images themselves, and the points are set according to «best guess».

The same applies to height above ground. The RO and the referees themselves have to guess how far it is to the ground under a player in the air. This is probably what happened when Moen claimed 35.5 cm offside. They completely missed where the ground was. This can therefore give quite a few centimeters of variation when the lines are set. Marginal situations can be seen as both onside and offside, with seemingly credible images in both cases.

This is not just a Norwegian issue though. Another recent example is from Crystal Palace – Leicester some days back, in a league with a lot more money to spend on VAR than our own. VAR forgets to estimate height above the ground on the attacker, but they did do it on the defender.

A quick check with more lines from @futoffsides reveals that this the attacker was most likely OFFSIDE. A correct decision by the match official was overturned by VAR.

Potential deviation points and height above the ground: ± 300 millimeters

When you have to set offside based on a shoulder, there is also a lot of room for variation. In many cases, this is placed from the armpit and downwards. But this can of course also be complicated by the shirt fluttering, or the point being set on another part of the shoulder – and that fact that it can be difficult to calculate the distance down to the ground from the point you set.

In a situation referred to as «4 cm offside», the margin of error is actually significantly greater than 4 cm itself. The picture below was taken during work and is not a finished result, but it nicely illustrates how small a point is in a large picture of the track.

The numbers are probably higher

The numbers above refer to the distance the body, foot and ball actually move, but in addition, you must remember that 2 pictures can draw a larger outline of a movement.

• The exposure in the first frame starts at 0 ms and continues to 20 ms
• 20 ms pause between images (or odd/even interlaced exposure in 50i)
• 20 new ms with exposure

In total, there will actually be 2 images with 60 ms exposure time on a foot in motion – corresponding to 864 mm from where a foot starts to draw in the first image to where it stops drawing in image no. 2.

The numbers in these calculations are therefore actually higher depending on the situation, but I try to keep it a bit simple. There is a headroom here of 50%, so the figures in the summary are in no way figures that have been inflated as much as possible.

Conclusion

The weak point in the chain is not the software at the bottom, but the limitations in the video images that are superimposed and the way these are interpreted by the VAR referees. The limitations in the images are known, or should have been known, also in 2020 when it was decided to investigate VAR. It is relatively simple math.

We already have the luxury edition of VAR.

Terje Hauge, head of refereeing NFF – VAR-debate at Nova Spektrum 5th of September 2024

Terje Hauge stated this just hours before this article was finished. VAR manager Moen emphasizes that this is not «First Price» (that would be Wish/Temu for you guys), but that it is used by UEFA. In other words, the two top managers of VAR in Norway have given a pretty clear signal that VAR technically cannot improve much.

It makes little difference that the same technology is basically used in UEFA, as the VAR manager tells VG. Norwegian VAR is never better than the weakest link in the chain.

High margin for error + low threshold for intervention = potentially many errors

VAR in Norway has operated with a stated expectation that the referees can judge with precision down to the millimetre. This is mathematically impossible to defend. In all situations described above, the margin of error is much larger than this. This means that marginal decisions where VAR has overturned the referee’s/match manager’s decision can just as easily be wrong, in some cases probably wrong and in some investigated cases clearly wrong.

Willingness/ability to accept feedback

Lack of investigation, training and willingness/ability to accept technical input has been central in match-deciding situations and overturned referee decisions in Eliteserien matches for over 1.5 seasons. At the same time, clubs and supporters have been met with a steadfast argument that VAR is millimetre-precise, that we are driven by emotion, and that we have not familiarized ourselves well enough with how the system works.

The head referee and the VAR officer have come across as self-confident and unwavering in their stance and their opinion that they know the system, no one else.

It has taken weeks to admit completely obvious mistakes, and a year and a half to acknowledge that you cannot measure in millimetres. When this realization came it was wrapped up in «we’re still right, but we’re making a change from now on».

Huge variety in practical use

In two specific examples, the VAR room’s assessment has changed radically. When Brann was approved 1-0 against Tromsø in 2023, the first measurement was 20 cm, and the second measurement 18 millimeters. VAR therefore missed by 11x on the first attempt. In the example from KFUM-Glimt, the first measurement was stated to be 35.5 cm, and the next 5 cm. A reduction of 85% of the original «not even marginal» offside, that is.

It is completely understandable if one has limited confidence also in the 2nd measurement, when the 1st measurement is so wrong.

Negative consequences for match officials on the pitch

In offside cases, the referee does not look at the screen himself, and relies on messages from the VAR room («In-ear facts»). This can understandably create some confusion and negative learning, as referees who are present at the match may have made a completely correct assessment and then be told they were wrong.

The conclusion is thus briefly summarized:

1. VAR in Norway is already a «luxury version» that cannot be improved

2. The only way to «improve» VAR is to spend more money on TV production in the Eliteserien

The only ones who notice anything in particular about this luxury are probably the VAR referees, who get NOK 9,563 and NOK 5,983 in fees without even breaking sweat. It is also no disadvantage for TV2 that there will be more/more expensive cameras in the arenas, although they are unlikely to pay for the party. The only question is who will pick up the tab. Levanger and Åsane? (Small clubs who have never and will never play with VAR)

Reflection: So what does it take to «improve VAR»?

Many of the clubs who are in favor of improving VAR also do not say much more than that it must be «better». An underlying consequence in that case is that VAR will also be much, much more expensive – because the next logical place to improve is to raise the quality of the video images.

Let’s say you upgrade all the offside cameras to shoot 100 frames per second and only get 10 milliseconds between each frame. One can then reduce the deviations for a ball in speed to only move 279 mm between each frame. A moving foot with a shutter speed of 1/200 second will be able to move 144mm between two images.

In a full round of games, there will then be 16 much more advanced and more expensive cameras in use. And that is only the first link in the chain. It will also cost more in every step the images have to go through, and require much more storage.

Now I’m just going to take a completely theoretical example based on equipment I usually use. A «normal» recording in XAVC-I in full HD uses approximately 112 mbit/s, or 14 megabytes per second. 90 minutes + 5 minutes of overtime then corresponds to 79.8 GB per camera. If you shoot 100 FPS, it will (slightly depending on the infrastructure in the chain) quadruple. And since you must have an offside camera every 16 metres, it is doubled once more. Multiply that by 8 games. Multiply that with 30 rounds. Year after year.

It may be that a solution is found to handle 153,216 gigabytes only from offside cameras, for example that material is deleted continuously, or that a solution is developed that only caches offside situations/VAR situations.

Can we use AI?

For fun, I ran the situation from KFUM – Glimt through Topaz Video AI, a widely used tool in my industry. Here, the clip is slowed down to 1/8 of the original speed, using 8 AI-generated images per original image.

Although in some situations it can provide a smoother movement of objects at low speed, it introduces a new set of artifacts and distorts reality. AI slow mo may be stylish, but it has no place in VAR.

Recently rumours will have it that an AI solution to offside decisions will be launched. Just remember that the AI will have to work with the same video foundation as now, and it learns from… us.

Improving VAR only means more costs for the TV broadcasts

My point is that probably very few people really understand the options, scope or cost of what it means to «improve VAR». Now I’m not going to do the NFF and estimate 200 million out of thin air, but there are potentially multiple millions every year, which the FA has to pay. And the NFF’s profit of 5.2 million in 2023 may run away on quick feet, feet that move much faster than 14,400 millimeters per second.

Do we have this money? Is it worth spending this money on VAR instead of, for example, player development or other things?

The Sunday league clubs is perhaps somewhat cut off now from deciding whether we will use VAR or not. But if the FA want to spend several million extra a year on VAR, they might object.

One thing is certain; it costs much less to say that you want to improve VAR, than to actually improve VAR. And Levanger, for example, should think about that before they have a board meeting where they decide that they want to keep and improve VAR in the Eliteserien.

Those who are in favor of keeping and improving VAR must present a reasonable cost estimate for how much more this will actually cost, not just scare the other side with the fact that it may cost more money to abolish it – an unfathomable claim in itself.

If not, the vote must consist of abolishing VAR, or keeping VAR as it is today. Anything else would be to mislead the Football Parliament.

About the author

Bjørnar Posse Sandboe is a resource person and former spokesperson at the National Supporters Alliance, chairman of Stabæk Support and an opponent of VAR. I have worked with technology and TV for 17 years. The last 13-14 years with large and small programs on all Norwegian channels and several international streaming services, many of them as DoP. It gives a certain experience, and it gives a humility that there is always more to learn. A hammer may be a good hammer, but it only works as a good hammer if you know how to use it.

I have been in the VAR room on 08/22/2024 and seen the technology in use together with representatives of TV2, NFF, EVS (supplier) and DCM. Some of the images are from here, and they are work in progress and not the exact same ones that were used when the matches were played.

What I write above is hard science to the best of my experience and can be challenged by anyone. I have spoken out about many, many things about the VAR process, and was early on to warn the head referee against believing that this system is millimeter precise as it is used here. It fell on deaf ears for over a year, but recently VAR has acknowledged that they do not have enough accuracy to rule on marginal offsides.

We are left with a large number of marginal decisions that may be completely wrong, which nevertheless count as «correct VAR interventions» in the NFF’s figures. It is important to be aware of this when you are presented with these figures as facts.