Use of technology in figure skating tech calls

[draqq]

I actually think under-rotations should be de-emphasized in the sport. I think that this kind of technology would only make these decisions more important than the should be.

Determining under-rotation is just too finicky and subject to bias when there's 1-3 people determining whether it's been under-rotated and by how much. I would rather have the system alert judges as to whether there might be an under-rotation (similar to a ! on an edge call) and then having the panel of judges determine the GOE penalty as they see fit. If it's clearly a < and <<, then the technical panel can mark it as such, but give some flexibility here.

Also, under-rotating a quad toe by 90 degrees but otherwise standing up is lower in points (7.2 BV - 2.0 GOE = 5.2) than a fully rotated quad toe with a fall (10.3 BV - 3.0 GOE - 1.0 Fall = 6.3). How does that make any sense? In fact, many judges give -3 for under-rotation too so the difference in points could be even wider.

And are you also telling me that someone who rotates a jump 80 degrees short deserves 3 BV points higher than a jump 100 degrees short? And then does the landing count when the full blade touches the ice? Or is it when the toe pick lands on the ice? Or does the skater's back position matter? Again, this is way too finicky of a procedure for something that can easily determine whether one skater wins or loses a podium placement.

 

[Vanshilar]

If you're going to call me a liar, do it outright, please.

I don't know why you like to imply motives so easily, when you asked those questions but didn't give your own experience, I was asking if you had personal experience, and I directly said I was looking for someone who could personally vouch for what the tech panel does, because I want some info on its actual practice (as opposed to just what can be gleaned from the rulebook etc.).

First, every skater who stands up on a jump has landed on their toe to some degree. If you land on a perfectly flat blade, you're going down without some serious luck and strength. But even a very "toe heavy" skater still does not land solely on their toepick. And yes, it is the point of contact with the ice that is examined. You would be amazed how many people do "hook turns".

It is not impossible to rotate a jump if you point your toes. Jason Brown rotates the majority of his and he has very pointed toes. Plushenko points his toes well too. He's never had a problem with URs. It is purely a technique problem if you cannot rotate jumps - it has nothing to do with how pointed your toes are.

Certainly, if you're good enough (jump high enough, rotate fast enough) that you have no problems with under-rotations, then it doesn't really matter how pointed your toes are. But if you tend to be borderline on rotation, back-of-the-envelope calculation gives that fully pointed versus fully flat feet can mean a difference of up to about 40-50 degrees of rotation (calculations at bottom), although in practice it'd probably be more like 20-30 degrees or something, which may make the difference between getting a UR call versus not getting one. Basically, it can be significant -- which also means that people who are borderline but also tend to point their toes more than other skaters are somewhat at a disadvantage (I mentioned Mirai because she was already cited earlier in this thread). That's a consequence of measuring the rotation from the instant the toe-pick touches the ice as opposed to when the blade proper touches. However, if that's the standard, and skaters are aware of it, then it's their own judgment as to how much they want to point their toes (knowing that more flat means harder to control but less likely to get UR call if they're borderline)...I guess just one of the many tradeoffs as they strategize their execution.

Fine. I have personally seen it being used. I have been the one in control of the buttons. I have sat on a flight of 40 skaters and you could probably count on one hand the number of times we didn't use it.

Cool. I'll cite this then when I (eventually) write my post about determining jump rotation. (I have a list of citations for various aspects of jumps, but not specifically for how the tech panel does the determination.)

It varies from panel to panel. TECHNICAL PANELS ARE MADE OF HUMAN BEINGS. Some are naturally more lenient than others - even when they think they're being harsh. It also varies from level to level. At a lower level, a technical panel might be a little more lenient. If, after multiple looks on the slo-mo, the panel can't agree, benefit of the doubt usually goes to the skater.

I was hoping for a specific figure, like "10 degrees" or "30 degrees" or whatever. Yes, it may vary depending on panel, but that's actually a problem with human judging, the inconsistency. Yes, they may be more lenient at lower levels, but the focus here is more on the elite levels of competition, i.e. Worlds, Olympics, etc. Basically, how much leeway is generally acceptable in practice? Although, I suspect the true answer to this question won't be from estimates but from looking at the actual data (comparing jump rotations to UR calls), although I've already mentioned 3 cases that (it would seem to me anyway) shouldn't be acceptable as not getting UR calls if the judges were diligently using slow-mo and all that to gauge under-rotation (I mean, if some random internet guy can do it...). In which case, it's probably better to ask, how much should be acceptable?

There's a reason for framing it in this way. Currently, the tech panel basically decides UR or not, essentially a binary choice (ignoring downgrades and such). From a value perspective, it's essentially saying "80 degrees under-rotated is okay, but 100 degrees is bad and equally bad as 170 degrees under-rotated". The first problem with this is that for borderline calls, it forces the tech panel to make a fine distinction that significantly affects the scores. The second problem is that it doesn't actually represent how people actually value the under-rotations -- people likely think 100 degrees is slightly worse than 80 degrees, but 170 degrees under-rotated is significantly worse than 100 degrees, but the scoring system doesn't take this into account.

If the amount of under-rotation could be numerically quantified, however, then the scoring system could be more like the timing of races, i.e. on a more graduated basis. For example, the scoring could be something like no penalty for 90 degrees or less, but a 1-point deduction for every 30 degrees of under-rotation beyond 90 degrees, linearly interpolated (so 135 degrees would mean a 1.5-point deduction). Then the scoring wouldn't be so much a luck of the draw by the tech panel for borderline cases, but more reflective of the quality of the jump (with regards to under-rotation). It would also eliminate the Internet caterwauling about how judges seem to be more lenient on some or other skaters that are borderline, since a UR call or not can mean several points that decide a competition.

Remember, too, the panel is on a time limit. After 10 minutes the remaining skaters in the warmup group have to go on for another six-minute warmup. Spectators, coaches, skaters and their families all get quite cross with you if it takes too long. The record, I think, for the longest review period I have sat on was eight minutes - a horrible affair in which a nine-element program had eight reviews. The only thing not reviewed was the step sequence. On the flip side, if a caller has called a skater in real time and there are no reviews, it's quite possible to have it done in as little as 20 seconds.

There's several comments to this. One, if the tech panel is rushing through the determinations, then (depending on how you look at it) it points to the inefficiency of the process and/or the competence of the judges if even at Olympic-level competitions they'll flub the calls quite often because they're rushing through it. Is this the type of officiating we want for this sport? I think that's getting more at what the OP was saying.

Two, automating some or all of this process would speed up the judging. For example, determining the skater's X-Y position can be automated relatively easily, and with the "close-up" camera (in this case, the one the tech panel uses) having a known position relative to the rink, the orientation of the camera relative to the path of the skater can be calculated automatically. The remaining issue then is the orientation of the skate relative to the camera. There are a variety of techniques for this, but I'll give a simple one (it's the one I use to do a relatively quick approximation). Assume that the skate's rotation is fairly constant in the latter part of the jump, up to the landing. Look for the frame when the skate is pointing directly at the camera (so the blade is close to a thin line), and count the number of frames until the skate is pointing directly away from the camera (so the blade is close to a thin line again). This means that in this number of frames, the skate rotated 180 degrees. Assume this rotation rate continues until the skate touches the ice. Now it's more likely that the skate's rotation slows down, but this has the effect of undercounting the under-rotation (acting to the benefit of the skater), which is fine. This will give the angle offset with respect to the close-up camera. Add this to the angle between the close-up camera and the path of travel, and you'll have the estimated amount of under-rotation. Basically, instead of eyeballing the landing by forward and reverse slow-mo, while trying to mentally adjust for whether the skater was moving somewhat toward or away from the camera at the time, the tech panel just needs to do some counting of the frames and add it to the angle offset calculated by the positions. It's a faster way of determining under-rotations. Of course, ideally, software algorithms will directly recognize the orientation of the skate relative to the camera, but image recognition for this may or may not be tricky (I have my own thoughts on how to do this, but haven't carried them out), so this is if it's still humans doing the adjudicating.


As a side note, although I mentioned a two-camera system, it's possible to do all this with a single "close-up" camera, provided that the camera setup can record its angle (both vertically and horizontally), its position relative to the rink is known or calibrated, and its zoom amount is also recorded. To calibrate the camera would just mean pointing it directly at different known positions of the rink (along one edge, then another edge) and recording its angle for each of them. Then, where on the rink it's facing can be calculated from its angle. The zoom level would then give the pixel size, i.e. the equivalent rink distance for each pixel. In this way the skater's position can be calculated from the close-up camera (that follows the skater around) alone without needing a side camera that's looking at the entire rink. The drawback is that the close-up camera had better not move from its stand for the entirety of the competition (or at least, that it has to be recalibrated if it does move). In fact there are many ways the jump parameters (rotation, height, etc.) can be done by video; the process I usually describe is more geared toward the two-camera approach (one wide, one close-up) because that's the type of data I can find online.


Back-of-the-envelope calculation: I wear male size 8 boots (not particularly big). I measured the reach difference between fully flat and fully pointed foot to be 8 cm. Say a jump has an air time of 0.6 seconds (corresponding to a height of 44 cm). This means on the way down, it accelerates for 0.3 seconds, so the velocity at the ice will be around 294 cm/s. The 8 cm will be covered at this velocity in about 0.027 seconds (yes, the skater will still be accelerating, so it's actually somewhat less). Assume a skater rotates at around 5 times per second in the air, or 1800 degrees per second. In that 0.027 seconds, this means the skater would have rotated about 49 degrees. In reality the skater would have kept accelerating so the rotation would be less than this, plus the correct comparison is not between fully flat and fully pointed but "more flat" versus "more pointed", but this shows that the amount of rotation can be significant.

 

[ice coverage]

The cool blade-cam video that Andrew Dodds posted a week ago seems worth adding to this thread.

https://instagram.com/p/7ULoJiDPz_/​

Also seems worth mentioning the "smart" blade outfitted with sensors that has been the subject of ongoing research (by a professor in Ithaca, IIRC?), as discussed in some previous posts/threads on GS.

 

[gkelly]

Would it be necessary to have 1) a "close-up" camera that focuses closely on the blades, for determinations jump takeoffs and landings (we sometimes see these views in broadcast replays, but those are not the cameras that the tech panel uses), and heck, maybe to settle disputes about entry and exit edges of turns in step sequences*, and also 2) a medium-close camera that shows the whole body, necessary for determining position-related features in spins?

*E.g., I recall that BoP and I disagreed about whether a certain unclear turn in Sotnikova's Sochi step sequence was a counter or a bracket, which made a difference in whether we thought she had achieved the necessary complexity of turns to earn level 4 for the sequence.

Or can the camera operators be trained to show the full body for some kinds of elements and close up on the blades for other types?

Would moves like dance lifts ever need to zoom in on the blades to confirm correct edges, in addition to looking at what's happening in both skaters' bodies above the ice?

How are official camera operators trained for ISU events? At club competitions it seems usually to be the same videographer who produces videos for the skaters to buy of their own performances, and thus it's all in medium shot showing the whole body.

 

[moriel]

Would it be necessary to have 1) a "close-up" camera that focuses closely on the blades, for determinations jump takeoffs and landings (we sometimes see these views in broadcast replays, but those are not the cameras that the tech panel uses), and heck, maybe to settle disputes about entry and exit edges of turns in step sequences*, and also 2) a medium-close camera that shows the whole body, necessary for determining position-related features in spins?

*E.g., I recall that BoP and I disagreed about whether a certain unclear turn in Sotnikova's Sochi step sequence was a counter or a bracket, which made a difference in whether we thought she had achieved the necessary complexity of turns to earn level 4 for the sequence.

Or can the camera operators be trained to show the full body for some kinds of elements and close up on the blades for other types?

Would moves like dance lifts ever need to zoom in on the blades to confirm correct edges, in addition to looking at what's happening in both skaters' bodies above the ice?

How are official camera operators trained for ISU events? At club competitions it seems usually to be the same videographer who produces videos for the skaters to buy of their own performances, and thus it's all in medium shot showing the whole body.

Btw, some motion sense cameras would be most likely possible. Shall be easy, the rink is empty, there is only the skater moving around, so maybe a few of them just following the skater from different angles could be quite useful, as would allow to cover many angles etc without the need of human operators.

 

[Vanshilar]

Would it be necessary to have 1) a "close-up" camera that focuses closely on the blades, for determinations jump takeoffs and landings (we sometimes see these views in broadcast replays, but those are not the cameras that the tech panel uses), and heck, maybe to settle disputes about entry and exit edges of turns in step sequences*, and also 2) a medium-close camera that shows the whole body, necessary for determining position-related features in spins?

I'll have to check if I can find any examples, but I suspect that if the "close-up" camera covered the whole body, it probably would still be good enough to determine jump rotations. The main issue is the frame rate of the camera. It needs to be high so that each frame is very clear and not blurry from the motion, for an accurate determination of the skate's orientation as well the precise moment it left or hit the ice. If it records at 25 FPS, and a skater rotates at 5 rotations per second (1800 degrees per second), then each frame would capture 1800/25 = 72 degrees of rotation. That's no good for determining jump rotation. If it records instead at 150 FPS (which I think was what the network close-up camera ran at), then each frame would capture 1800/150 = 12 degrees of rotation. For example, for Mao's Sochi LP 3A, the video I got ran at 25 FPS for the real-time performance. On replay, they showed the 3A in slow-mo at roughly 75 FPS (~24 degrees per frame), which is already enough to see that it was under-rotated; then they showed a close-up of the skates at about 150 FPS (~12 degrees per frame) which also made it obvious -- but the 75 FPS was already enough in this case. I'd be interested in seeing if I can find any examples from the Sochi video where they used the 150 FPS slow-mo of the whole body instead of just the skates, but that'll take some looking. (Hopefully they did for at least one of the 24 skaters of the long program).

If the frame rate is the same, the difference between focusing on just the skates instead of the whole body is increased resolution of the skates and surrounding ice, which is useful for determining the exact point of lift-off and landing, especially if there was ice spray. But I suspect that something like 150 FPS on the whole body would still have enough resolution of the skates to give an accurate measurement.

To give you an idea of how sharp the image is, from the network feeds of Mao's 3A:

From real-time video (25 FPS): http://s898.photobucket.com/user/Vanshilar/media/Mao 3A realtime_zpskqpplnxo.png.html
From full-body slow motion (~75 FPS): http://s898.photobucket.com/user/Vanshilar/media/Mao 3A replay full body_zpshtgyredb.png.html
From skates-only slow motion (~150 FPS): http://s898.photobucket.com/user/Vanshilar/media/Mao 3A replay skates_zpseeai6vsw.png.html (different vantage point but I selected where the skates are about the same angle)
Those 3 images side-by-side: http://s898.photobucket.com/user/Vanshilar/media/Mao 3A side-by-side_zps7kkngzl1.png.html

Keep in mind the close-up of the boots is not only ~2 times larger, but at twice the frame rate, so it's doubly-clear. To sort of simulate what it would look like at lower resolution, I made it half the size, at the right bottom of the side-by-side, but I think in reality (if it had been focused on the whole body) it would probably have been slightly more blurry than what the resized image suggests.

So TL;DR -- I think the important thing is the frame rate of the camera, so I suspect slow-motion of the whole body should be fine for determining jump rotation, as long as the frame rate is high enough.

I'm not sure how good algorithms are these days at tracking moving objects -- I *want* to say that the cameras for things like rocket lift-offs are still moved and zoomed by hand, but I could be wrong on this. The camera operator would need to be trained to keep the zoom such that the skates are always visible, though; one of my annoyances is that sometimes (especially when the skater got close to the camera) the camera operator didn't zoom out fast enough, so the skates were chopped off during some jumps.

 

[solani]

@Vanshilar
I really think it isn't that easy, you would need special equipment. The analysis has to be finished within 10min. and it has to be reliable. In my opinion (I repeat myself) you need motion capturing technique for that and I wouldn't stop with jumps there, you could use this for spins and step sequences as well. But I think it's technically not possible (yet). (If you don't want skaters in black suits with colored dots on their limbs and skates.)
I don't think that the ISU would be willing to invest money in such a project. It could get rather expensive, the rinks would need to be equipped.
And I think that the tech panels in general are doing a good job, not perfect, but good enough. Not everything is measurable when it comes to figure skating.

 

[jinhamasaki]

I'll have to check if I can find any examples, but I suspect that if the "close-up" camera covered the whole body, it probably would still be good enough to determine jump rotations. The main issue is the frame rate of the camera. It needs to be high so that each frame is very clear and not blurry from the motion, for an accurate determination of the skate's orientation as well the precise moment it left or hit the ice.

Asada participated in a competition where they showed the replays of the jumps after the program. The replay was a super slo-mo and was able to "rotate" around the skater to see the multiple different angles. I can't recall the competiton (nor find it) but maybe you know what I am talking about?

 

[Vanshilar]

@Vanshilar
I really think it isn't that easy, you would need special equipment. The analysis has to be finished within 10min. and it has to be reliable. In my opinion (I repeat myself) you need motion capturing technique for that and I wouldn't stop with jumps there, you could use this for spins and step sequences as well. But I think it's technically not possible (yet). (If you don't want skaters in black suits with colored dots on their limbs and skates.)
I don't think that the ISU would be willing to invest money in such a project. It could get rather expensive, the rinks would need to be equipped.
And I think that the tech panels in general are doing a good job, not perfect, but good enough. Not everything is measurable when it comes to figure skating.

Well, how easy it is will be shown once I make the post, but I can already do it manually (I input stuff into Excel and I have some of the equations put in, but I haven't written it into an automated program yet). My claim is that you just need a couple of video cameras (one that captures at a relatively high frame rate) and software. Note that I'm not talking about measuring the position of the entire body, but more or less just focusing on the skates (i.e. jump height is from when the skates leave and then contact the ice, jump rotation and correct edge is measuring the orientation of the skates on take-off and landing), in which case the focus is on the bare metal of the skates. If it's automated by software, the analysis could be done in real-time as the performance is in progress, so (ideally) it means that the judges will already know the measurements (such as "under-rotation = 107 degrees, edge = 5 degrees on inside edge) by the time they do their scoring.

The equipment is not the hard part (other than the video cameras), the hard part is the software algorithms that needs to be developed. The difficulty is that we as humans are intuitively good at looking at video and recognizing things like feet and knees and stuff, but it's difficult to teach a computer to recognize them (which really means, write algorithms that will recognize them). Thus my focus is on what can be done relatively easily and then moving to the harder stuff. The goal isn't to replace the entirety of judging by computers, the goal is to find what is easier for a computer to do and have the computer do those things so that 1) they can have more consistent results, especially for quantifiable measurements like jump rotation and 2) humans don't have to waste their time on it. Stuff like how good a spin position is will likely still be done by human judges for a long time (although I think how centered a spin is may be computerized, depending on how often a skater changes position).

It may be possible to measure the position of the entire body from video, but I don't know if it's technically feasible at this point. I know there's a lot of research into things like gait analysis (that scene from Mission Impossible is based on real tech, just not so...convenient), so in the future it could be applied to things like figure skating, but I think that's off in the future.

Note that I'm a proponent of using video image analysis because 1) I have some familiarity with it 2) it's something anybody can do 3) it can already be done from publicly available media (I just need to find the right Youtube videos basically) 4) it can be applied to past competitions (to compare with judging results) 5) other than the cameras, it doesn't require any special equipment (though software is the pain) 6) it is non-intrusive to the rink and to the skater. Yes you could attach sensors and stuff to the skates or to the rink, but I tend to think that those hardware makes skating more inaccessible and increases the costs logistics of a competition. For example, attaching sensors to the skates may make it more difficult for the skater to move (they add weight, after all), have the possibility of breaking from jumps and such, you would need them for each competitor (or for each group of competitors) and/or be switching them out really fast between competitors (which increases the possibility that one of them might fall off during a performance which would be very bad especially safety-wise), they'd have to be calibrated for each skater (whereas a video camera just needs to be calibrated once for the competition, as long as it's not moved -- to know where the camera is in relation to the rink), etc. So there are many other technological possibilities, but I tend to favor this one.

 

[solani]

My problem with your approach is simply that I think it will be unreliable and inaccurate. It's not that it wouldn't work 95% of the time, it probably will work very well in most of the cases. You will get different results from different camera angles. My point is simply that if you want to measure jumps with technical assistance you have to get it absolutely right. It must be as accurate as clocking the 100m sprint. A lot of testing would have to be done before that, with fixed cameras and you simply have to have a camera in bird eye's perspective because you have to know in which direction the skater is jumping. You cannot do this accurately with existing videos, because you don't know the position of the camera.

 

[karne]

While the use of more technology could be helpful, I don't understand the obsession with trying to eliminate the human factor. It's part of every single sport!

Let me think of some examples. In motorracing, if there is a collision, it will be referred to the stewards. Three different stewards might look at the same rulebook, the same camera angles, and come to three different conclusions. Steward 1 might say, "Driver A turned in on Driver B, therefore, Driver A's fault." Steward 2 might say, "Driver B threw himself recklessly up the inside where there was not room, Driver B's fault." And Steward 3 might say, "They are equally at fault. Racing incident."

In rugby league, there is a video referee, with super slo-mo and frame-by-frame to determine tries - and the video referees STILL get it wrong.

Technology is not the be-all and end-all solution; it will not automatically make it so that no judging scandal will ever occur, or that fans will never disagree with results.

 

[gkelly]

Judging skating programs is more precise with video replay than without (introduced in the mid-1990s), and with IJS-style technical panels than with just judges.

More accurate measurements of jump rotation and correct edges, maybe even of speed and other measurable data, could be more precise still -- when and if technology is readily available for such uses in ways that would be affordable beyond the top elite level of the sport and that don't interfere with the actual execution of the skating.

But the things that can be objectively measured are only part of what makes one skating performance "better" than another. As long as qualitative factors count, human judgment will be necessary and disagreements will be inevitable.

As for the technology side of it, it's interesting to explore what might be possible and how it could be used, and which technologies are already far enough developed to be implemented right now.

But until someone brings the technology into a rink and tests it out in a real competition setting, I don't think we can know whether it is ready yet or not. So it makes more sense to try to figure out what might work than to take the Powers That Be to task for not having already implemented something that might or might not already be able to serve the imagined purpose.

Thanks to whoever changed the title of this thread to focus the discussion in that direction.

And if the ISU is in fact looking into ways to use technology in ways that would change the scoring process yet again, we probably won't hear much about it until the details are ready to be implemented in real competitions. Probably first with a test at Nebelhorn.

 

[jinhamasaki]

Not even Hawkeye is 100% accurate. I read that it has about a 2 cm margin of error!

 

[Vanshilar]

My problem with your approach is simply that I think it will be unreliable and inaccurate. It's not that it wouldn't work 95% of the time, it probably will work very well in most of the cases. You will get different results from different camera angles. My point is simply that if you want to measure jumps with technical assistance you have to get it absolutely right. It must be as accurate as clocking the 100m sprint. A lot of testing would have to be done before that, with fixed cameras and you simply have to have a camera in bird eye's perspective because you have to know in which direction the skater is jumping. You cannot do this accurately with existing videos, because you don't know the position of the camera.

Well, consider the current judging system. Do you think the current judging is absolutely right? If not, then why does any change to the system have to be held to an "absolutely right" standard, if you don't hold the current system to the same standard? There's an old aphorism, "Perfect is the enemy of the good" -- the point is to improve the judging system, rather than do-or-die with a mythical "perfect" system.

The camera angles issue is an interesting one -- because the judges sit on one side of the rink, and (IIRC) the video camera used by the tech panel is also from that vantage point, so under the current judging system, they only have one vantage point to do their judging from already. So the problem you bring up already applies to the current system. The reason why I propose a "wide" camera that captures the skater's X-Y position within the rink is that by doing so, it more or less eliminates the camera angle issue (at least with respect to under-rotation), so the close-up camera only needs to determine the skate's orientation relative to itself. I *do* think that pretty much all the Internet videos I've seen that purport to show under-rotations pretty much ignores the camera angle issue which makes them bogus, but the "wide" camera produces the correction factor that eliminates this issue.

Regarding accuracy, how accurate do you think software algorithms need to be? More importantly, do you think that humans are more accurate? For example, from network slow-mo videos of 150 FPS, the accuracy of calculating jump rotation is around 15 degrees or so (the limitation is the FPS, higher FPS = more accurate). Do you think the current judging method (i.e. human eyeballs looking at slow-mo video) is more accurate than within 15 degrees?

Of course a lot of testing needs to be done prior to putting it to use, just with any other change. But part of the reason why I propose to do this using existing videos is because this allows it to be developed in parallel with existing methods. Not only that, but because existing videos are used, it allows for transparency as well as a backup in case people think the software algorithms aren't correct: they can look at the images themselves. I think this is better than something like sensors on the boots because it's then harder to present the data in a way that is easily digestible to judges/experts and/or the average layman.

I do in fact claim to be able to do this accurately with existing videos; not only that, I will show how to do this with existing videos (namely, videos of the Olympics), mainly because I don't have my own camera equipment nor the budget to buy them and travel to competitions. The fancams that I'll use in effect act as the bird eye's perspective that you mention because they show the entire rink as well as the rink-side cameras used for most of the network videos, so I'll know the position of the network cameras. You are right though that if the camera's position isn't known, it's very difficult (though not impossible) to do; but I target the Olympics for this specifically because there's a variety of fancams of it that I've already found so I can figure out the camera positions.

While the use of more technology could be helpful, I don't understand the obsession with trying to eliminate the human factor. It's part of every single sport!

I think that's a fundamental misunderstanding that people have about technology in general. The purpose of technology isn't to replace humans. I think people think of technology as if we'll be competing with robot Terminators that will be flipping burgers so they can earn enough money to buy lubricant oil (yes, I'm aware that burger-making robots have been invented already). Technology is used to change how people do things, but the impetus is inherently still humans. In this case, it's a question of whether software algorithms developed by humans to analyze videos can be as good or better than people peering at videos themselves. There's also the "intermediate" alternative of whether humans analyzing videos in a different way is better as well; in fact, I favor this approach until the development of better software algorithms. This is pretty much like using rulers to measure distances and scales to measure weights: because for those measurable quantities, technological devices are more accurate and more consistent than humans.

Let me think of some examples. In motorracing, if there is a collision, it will be referred to the stewards. Three different stewards might look at the same rulebook, the same camera angles, and come to three different conclusions. Steward 1 might say, "Driver A turned in on Driver B, therefore, Driver A's fault." Steward 2 might say, "Driver B threw himself recklessly up the inside where there was not room, Driver B's fault." And Steward 3 might say, "They are equally at fault. Racing incident."

In rugby league, there is a video referee, with super slo-mo and frame-by-frame to determine tries - and the video referees STILL get it wrong.

I don't know if you noticed, but in the examples that you gave, it's humans that get it wrong. That different humans look at the same situation and come to different conclusions means that humans are inconsistent at gauging it. In your second example, how do you know the video referees get it wrong? It's because you the viewer (human) judge the same video differently than the video referees (humans) -- in other words, different people made different judgment calls.

If you wanted to find out how heavy something is, would you rather hand to someone to hold and tell them to give you a value, or would you rather put it on a scale? Certainly the scale may give a wrong value if it's not properly calibrated, but that's what technology does; it gives people a different role, not replace them. In this case, the human calibrates the scale, then the scale measures, and this process is more accurate and consistent than having people hold something and give their intuitive estimate. The same goes for video analysis.

Technology is not the be-all and end-all solution; it will not automatically make it so that no judging scandal will ever occur, or that fans will never disagree with results.

No, technology is meant to improve judging, not to replace it. For example, in those examples that you gave, could you say "well see if we *didn't* have this technology, we would've gotten the call correctly"? For example, does someone who doesn't have access to slow-mo make better calls than someone who does have such access? Nobody said technology is some magic wand that will make the world perfect and give us unicorns and rainbows; it's merely a tool to assist us in our endeavors.

With better measurement technology, it doesn't mean that fans will never disagree with the results, but it will mean that fans will have much less disagreement about the measurable aspects of skating, such as under-rotations. (Unless fans propose better analysis of the videos or something, in which case I welcome it.)

But the things that can be objectively measured are only part of what makes one skating performance "better" than another. As long as qualitative factors count, human judgment will be necessary and disagreements will be inevitable.

Incidentally, I posted about this about a year ago when there was another thread about using better measurement technology in skating:

http://www.goldenskate.com/forum/sh...kating-Imagine&p=885727&viewfull=1#post885727

The gist is that technology can only measure ("how much"), but they can not give value ("how good"). In the case of under-rotations, technology can only give a calculation of the rotation angle, it cannot score that angle without human input. So even for quantitative factors, it's still humans controlling the process, it's just that humans don't have to go through the tedium of doing the measurements (or at least, make the tedium more streamlined if the process isn't fully automated). For qualitative factors, well, it's humans all the way until they are reduced to quantitative factors, and technology can only assist (i.e. slow-motion video to help humans gauge the artistic quality of a spin position).

As for the technology side of it, it's interesting to explore what might be possible and how it could be used, and which technologies are already far enough developed to be implemented right now.

But until someone brings the technology into a rink and tests it out in a real competition setting, I don't think we can know whether it is ready yet or not. So it makes more sense to try to figure out what might work than to take the Powers That Be to task for not having already implemented something that might or might not already be able to serve the imagined purpose.

Yeah I think that's part of the motivation for analyzing the videos -- if some random internet guy can do it, certainly people that are specialists in the field (image processing) can do a much better job of it. It's to show that the method can work, as a proof-of-concept. I certainly don't intend to write the software or start my own company analyzing skating metrics (I have some background in image processing but am far from an expert, I don't have the time because I"m working on my dissertation, this is a side interest not something I'm going to make a career of, etc.), but to show that we can already do a much better job of analyzing these measurable quantities than the current method of eyeballing things. Additionally, I think it'll lead to much less disagreement about what occurred; if there's an objective process for doing the analysis, then it's difficult for people to argue back and forth "the video showed it was under-rotated!" "no, the video showed it was not under-rotated!" which doesn't really elevate the discussion nor move the discussion forward.

 

[solani]

Well, consider the current judging system. Do you think the current judging is absolutely right? If not, then why does any change to the system have to be held to an "absolutely right" standard, if you don't hold the current system to the same standard? There's an old aphorism, "Perfect is the enemy of the good" -- the point is to improve the judging system, rather than do-or-die with a mythical "perfect" system.

An automatic measurement has to be absolutely right, I hold any automatic system to that standard. If machines do our work they cannot fail. It's as simple as that. And we know many automatic systems that need interference from human operators at certain points and that's what you don't seem to want for your system. I would have nothing against a technical system like the one you suggest as an assisting tool for the technical panel.

 

[Vanshilar]

I actually think under-rotations should be de-emphasized in the sport. I think that this kind of technology would only make these decisions more important than the should be.

I actually have the opposite view -- that more accurate measurement of under-rotations would actually result in de-emphasizing it in the long run.

Currently, the way under-rotations are measured is for people to eyeball it in video playback. It's difficult for humans quantify things accurately by eyeball, so under-rotations is simply divided into ranges: 0 to 90 degrees is "okay" (no deductions), 90 to 180 degrees is "under-rotated" (some deduction), above 180 degrees is "downgraded" (heavy deduction). Even though I agree that 100 degrees is not much worse than 80 degrees under-rotated, the IJS system considers 80 degrees to be just fine while 100 degrees to be just as bad as 170 degrees, both worthy of a sizable penalty.

If under-rotations can be more accurately measured, however, then it becomes feasible to change the points system to reflect the more accurate measurements -- where 100 degrees under-rotated is a small penalty, 120 degrees under-rotated is a somewhat bigger penalty, 150 degrees under-rotated is a yet bigger penalty, etc. That's why my suggestion is to simply have it be a linear scale if we had better measurements. In this case, provided the measurement is relatively accurate (i.e. if most people will agree that the actual amount was within say 10-20 degrees of the measured value), it wouldn't be worth arguing over, because the points difference between 80 degrees versus 100 degrees, or 100 degrees versus 120 degrees, would be so small. Right now, under-rotations is worth arguing over precisely because the points difference is so big. Thus, with more accurate measurement, under-rotations will be de-emphasized, at least as a point of contention and Internet drama.

Does this mean that under-rotations become unimportant? Not at all. If a skater chronically under-rotates by around 135 degrees, then he/she can pretty much expect that amount of points to be deducted from the performance. So there remains an incentive to not under-rotate; chronic under-rotaters will consistently lose a certain amount of points if they don't improve. However, this reduces the variability and unpredictability in judging, where sometimes people will under-rotate by 135 degrees and not be penalized while others under-rotate by 120 degrees and get penalized, etc., because it's currently done by eyeball, and it's a guessing game for the rest of us as to how the judges will call some of the jumps.

Also, it allows the judging process to be much more transparent. Estimating by eyeball is inherently a very opaque process; it's hard to really examine someone's internal process for coming up with an estimate. By contrast, measuring by video analysis is much more open because everyone can observe the same data (i.e. images) being used, with a quantifiable justification for the numbers produced.

An automatic measurement has to be absolutely right, I hold any automatic system to that standard. If machines do our work they cannot fail. It's as simple as that. And we know many automatic systems that need interference from human operators at certain points and that's what you don't seem to want for your system. I would have nothing against a technical system like the one you suggest as an assisting tool for the technical panel.

So does this mean that you never visit any website, in case it might be down or you might lose a packet? That you never drive a car, because it might break down? That you're against traffic lights, because sometimes they break down? That you never use a cell phone, because it might not get reception? A lot of what we do nowadays is already automated to a large extent, so I don't know why you hold this particular case to a standard of "absolutely right", when the alternative that is currently used is anything but. Nowhere do I say that having an automated system means that humans can't interfere or are no longer involved, as if having software algorithms processing images suddenly means that humans aren't allowed to double-check the images as a sanity check to make sure the measurements were done correctly. (When I worked at a factory, even though the automated visual measurement machines were correct over 99% of the time, we still had operators who looked at each image individually to make sure the machines measured each image correctly.) It is almost certain that even if software video analysis can automatically measure under-rotation and so forth, there will be someone (likely the data operator) who does a sanity double-check of the images and measurement data, and flags possible bad measurements for manual review.


BTW although automated software analysis is the goal, what I'd be posting about is a lot more prosaic: how to manually measure jump characteristics (such as under-rotations) from videos. This is for several reasons; one, because I'll be posting about the details of the process, it will be open and something that anybody can do (provided they can find the right videos); two, so that there's a better understanding of how to extract such information from videos; and three, to have a firmer basis for discussion about the various events, instead of people just arguing back and forth "yes it was under-rotated" "no it wasn't under-rotated" without any actual resolution. To actually convert this process into an automated process (i.e. done by software), there would need to be software written to do things like image recognition of the skates, which is something I have ideas for but is not my specialty and far from something I'd want to write software for. But it would focus the problems on a few specific issues and show that measuring these metrics is not as difficult nor as mystical as one may think.

 

[solani]

So does this mean that you never visit any website, in case it might be down or you might lose a packet? That you never drive a car, because it might break down? That you're against traffic lights, because sometimes they break down? That you never use a cell phone, because it might not get reception? A lot of what we do nowadays is already automated to a large extent, so I don't know why you hold this particular case to a standard of "absolutely right", when the alternative that is currently used is anything but. Nowhere do I say that having an automated system means that humans can't interfere or are no longer involved, as if having software algorithms processing images suddenly means that humans aren't allowed to double-check the images as a sanity check to make sure the measurements were done correctly. (When I worked at a factory, even though the automated visual measurement machines were correct over 99% of the time, we still had operators who looked at each image individually to make sure the machines measured each image correctly.) It is almost certain that even if software video analysis can automatically measure under-rotation and so forth, there will be someone (likely the data operator) who does a sanity double-check of the images and measurement data, and flags possible bad measurements for manual review.

Oh, don't twist my words! Important technical services have to have a very very low failure rate. They are not 100% failure-proof, but very close to that. And if the autopilot on the plane doesn't work properly the pilot takes over. If the traffic lights stop working the traffic laws have to be applied by the road users. So every time something doesn't work as it should a human takes over. And the systems I mentioned work fairly well, the system you have in mind will not be as reliable. Humans, the tech panel (not the data operator(!)) have to make the final decision.

 

[gkelly]

The camera angles issue is an interesting one -- because the judges sit on one side of the rink, and (IIRC) the video camera used by the tech panel is also from that vantage point, so under the current judging system, they only have one vantage point to do their judging from already. So the problem you bring up already applies to the current system. The reason why I propose a "wide" camera that captures the skater's X-Y position within the rink is that by doing so, it more or less eliminates the camera angle issue (at least with respect to under-rotation), so the close-up camera only needs to determine the skate's orientation relative to itself. I *do* think that pretty much all the Internet videos I've seen that purport to show under-rotations pretty much ignores the camera angle issue which makes them bogus, but the "wide" camera produces the correction factor that eliminates this issue.

Keep in mind that you can't achieve the same kind of wide angle in all rinks where figure skating competitions are achieved. Not all are held in large arenas.

Where would you place the official wide angle camera in this rink, for example?

It looks like there are 6 rows of bleachers plus a kind of concourse behind them, with a few pillars at the level of the concourse. So probably the center of the top row of bleachers, or next to a pillar and projecting in front of it, would be the best place for a camera to capture the whole rink. Would such a camera follow the skater around the rink or remain fixed at a wide angle that only shows the whole ice surface? And how wide would that angle need to be with the camera relative close to the rink?

As local rinks go, this is pretty large.

Sometimes a US regional or sectional competition, or a national championship for a small skating country, will be held in one that's even smaller, with maybe 3 or 4 rows of temporary bleachers on one side of the ice and no space at all behind the officials' stand.

Would it be possible to use the same video technology in these rinks as in large arenas.

BTW, at these rinks the hockey glass is usually removed for figure skating competitions. But there may be a built-in structure for the central box, where the music and announcer are for competitions -- I don't know what they're used for in hockey. That can present a sightline problem for the official camera positioned next to the tech panel, for elements in the far corner of the same side of the rink. A second camera at a different angle could improve fairness in that kind of setup. But where would you put it?

 

[YesWay]

Keep in mind that you can't achieve the same kind of wide angle in all rinks where figure skating competitions are achieved. Not all are held in large arenas.

But does any proposed new technology need to be in every rink?

We already see a range of hi-tech aids used for calls/judgments in many other sports.. but not at every venue and every competition. It's frequently only at the kind of "big" venues used for major championships and finals... where stakes are high... and fan outrage at any contentious call is at its most shrill...

Perhaps it's more realistic for figure skating to work the same way? Only consider installation at major venues... and only hold finals and major championships etc, at rinks that support the technology?

 

[gkelly]

If it makes a significant difference in the way programs are scored and/or the way skaters put together their programs, then it would need to be used universally.

For example, the "well-balanced program" rules under IJS are significantly different than under 6.0, and under both systems there were occasional changes from one year to the next, and during the transition there were problems with programs designed for one system being competed under the other.

If it's just using technology to provide more accurate decisions on the existing < and << calls, that should be no problem.
E.g., judges have video replay available to them (not just the tech panel) at some important events but not all events they judge; at some lesser events there is no video at all and the tech panel just has to rely on what they saw in real time.

If the use of this technology is going to change how jumps are scored with respect to their rotation, not only in what the tech panel needs to do but also in how the base value is determined in the computer program and in how the call is communicated to the judges, then the computer would need to be programmed differently for events with or without the technology. Then the question would be whether that could introduce more problems than it would solve. Without knowing the details, it's impossible to predict.