Do it !!!
The issue is that it takes a fair amount of time, since although the math is worked out, the actual measurements are done by my eyeballs. The issue isn't with the math (which is already worked out and I've already wrote up an Excel file that takes the pixel measurement inputs and automatically converts them into rink coordinates), but with determining where the different rink locations (rink edges, skater blades) are in the image. It would be faster (and more accurate) if I wrote up an edge detection algorithm, but that in itself takes time. I was going to do it for the Olympics discussions but they've since closed those threads down so it's not as high of a priority for me anymore.
Also, it depends on what fancam videos (and of what quality) I'm able to find. So far I've only found them for Yuna, Adelina, Carolina, and Yulia. So those are the only ones I'm able to do a full analysis of each jump. The rest, I may be able to analyze them via an alternate method, but it would only be for several specific jumps (the ones that had a replay or a different feed from a different camera angle).
How do you measure the distance traveled by watching a video?
I've given an outline of it elsewhere, but basically, fancam videos tend to lack zoom so they cover the entire rink (or at least, parts of all 4 walls). I assume that the rink used at the Olympics was an Olympics-sized (heh) rink of 30 meters by 60 meters. I also assume that the walls around the rink are of the same height and vertical (with obvious exceptions to make space for cameras, etc.). The rest is perspective geometry. If the tops of the walls are in the image, then it's basically just a matter of "lowering" that 30 m x 60 m plane down to the level of the rink (and at least 2 sides of the rink will always be available to do this, if the walls are visible), and then determining the location of the skater's blades relative to the rink. This method isn't (easily) doable with the network feeds because they're zoomed in on the skater -- hence the use of fancams.
If the camera's viewing angle is perpendicular to the skater's path, it's fairly easy, provided you know the rink's dimensions, and the distance between the hockey features on the ice (center, offside lines, icing lines, faceoff circles). Your estimate becomes less accurate if you have to adjust for cameta angle.
The problem is that the camera's viewing angle is practically
never perpendicular to the skater's path, while it will
almost always look as if the skater is skating perpendicular to the camera, due to foreshortening (and camera zoom): that distances along the line of sight appear short or "flattened" compared with distances across the line of sight.
Failure to take this into account is the main reason why pretty much all of the post-Olympic analysis of jump under-rotations etc. are bogus, including some of the most controversial/heavily discussed jumps. (The other main reason is not estimating blade angle relative to camera properly.) By using a video that can see the entire rink, however, the skater's direct x-y coordinates within the rink can be estimated. Generally it's within tens of centimeters (the size of an image pixel), depending on the resolution of the fancam camera, which is decent considering the camera is usually just an ordinary camera filming from probably nearly a hundred meters away. (It's also worthwhile to note that for most purposes what's important is the
relative error between skater boot locations, such as measuring the distance of jumps, which in this case can be smaller than the error in determining the coordinates of the rink in the image.)
Additionally, there was a center line on the rink that was visible in some videos, which splits the rink in half, and can also be used as a basis to determine locations (in this case, the rink itself is used directly, rather than the tops of the walls).
Interestingly, the cameras used for the network feeds are visible in some of the fancam videos, so the network camera's x-y coordinates can themselves be estimated. In other words, the skater's x-y coordinates and the network camera's x-y coordinates can be determined from a fancam video. This means that the camera angle of the network camera relative to the skater's line of travel can be determined fairly accurately and thus accounted for. By accounting for this, then, whether or not a skater did under-rotate a jump (or by how much) can be determined fairly accurately, depending on the camera video quality (how blurry the skates are). With few exceptions, when people looked
only at the network feed and didn't take this camera angle relative to the skater's line of travel into account, under-rotations were either exaggerated or diminished depending on whether the skater was moving towards or away from the camera. Hence a lot of the "evidence" for them was flat-out wrong -- or, at best, significantly inaccurate.
An alternate method is to note that the locations of the patterns on the rink walls can also be determined (an example of this is here:
http://4.bp.blogspot.com/-QsIslU1dHmA/UvbpNzTHF6I/AAAAAAAACco/_-V2u0qXYug/s1600/DSC_0154.JPG ). With the network cameras' x-y coordinates determined, if there are two different camera shots of the same jump or move, the skater's position (and hence camera's angle relative to skater's line of travel) can be triangulated. This is less preferable compared with the fancam method but it's doable, and is the method I use for if the fancam isn't available. But I can only do it if I have multiple camera feeds, which means having videos from different sources and/or if the replay was from a different angle.
I haven't worked on this much lately because the Olympic-related threads are closed so there's no particular venue to put analysis of them now, plus nowadays I'm working more on school stuff (i.e. dissertation-related research) rather than on this. The principles and math are worked out, it's mostly just a matter of actually manually measuring skater locations for a bunch of different jumps and skaters and writing up the process (and results). That takes a long time to do -- which is why I mentioned this was a long-term project. Maybe by next year's Worlds or something. But a number of measurements were already worked out so it was relatively easy to use them to answer this question (speed of skater).
