0
JohanW

Judging a WRW formation

Recommended Posts

So, how does one judge a wingsuit relative work formation, resulting in a binary outcome - we got it, or we did not.

The grid idea as used in Elsinore needs a lot of computers, time and people to implement (channelling Jeff here, not necessarily correctly) and, with current skills, allows for a maximum formation size of about a 16 way diamond. For bigger formations, say a 100-way wedge, it's still very usable for debriefing purposes, I think, but, again with current skills, you'll never get a completed formation. This need not be a bad thing; skills improve.

An alternative is to put a diamond of appropriate size over each participant and see if the formation blots out all the sky behind it (assuming ground-to-air). This allows for crooked, skewed formations, which may or may not not be desirable.

Then there is the method with everyone touching a grid, be it chessboard or diamond style. That way madness lies, I have to give Yuri he showed that, when taken to the extreme. And you have to account for extremes.

Or we could go back to requiring grips. And to a 5-way world record. Not discounting the possibility. But imagine the effects on the glide ratio and the safety of the formation. Yuri would have a stroke because of the first, and I would not be in it because of the second. There's things to be said for either of those. :P It's extreme, but it's clear cut.

You might allow grips on soft noodles. Extend them a foot from hands and feet and allow the formation to skew and breathe just a little. Then separate and deploy .. B| can you build a formation with feet noodles only? Actually, yes, you can, but I still worry about the effects on the safety of the formation. No idea to what size you could (safely, or at all) build with current skills.

A possibility I threw out in a private conversation earlier is requiring grips for official records (I have a hunch the FAI might accept that) and judging bigways subjectively. Score Elsinore 8.8 out of 10 and accept it'll never be an official record. Personally, I don't like it. But if it's the only way of flying a 100-way, it'll do. I like flying bigways, even if only for the pretty pictures.

Then there's Kallend's idea:
Quote

Design the formation as a grid, but allow homogeneous dilatational and deviatoric strains to get the best fit to the jumpers in the photo. Then calculate the RMS error in jumper position from the best fit grid. If less than 2m, you have a record.

I would actually really like to hear him explain in layman's terms what this means and just how easy this is. Calling John Kallend!

Discuss ..
Johan.
I am. I think.

Share this post


Link to post
Share on other sites
Quote

Then there's Kallend's idea:

Quote

Design the formation as a grid, but allow homogeneous dilatational and deviatoric strains to get the best fit to the jumpers in the photo. Then calculate the RMS error in jumper position from the best fit grid. If less than 2m, you have a record.

I would actually really like to hear him explain in layman's terms what this means and just how easy this is. Calling John Kallend!


When I first read his post I thought he was joking, but reading it again I think I know what he means: figure out how far each wingsuiter was from his intended place in the formation and if the average deviation is less than 2 meters then the formation is considered complete. I quite like it, although we'd have to play with the deviation value and see what works and then finally agree on what defines complete is and what not. Another problem I see with this is that the judging becomes quite difficult, it is subject to human error, as you'd have to place each dot on the jumper in the picture, the location of which is open to interpretation (do you place it on the head or on the body etc etc).

I've been working on some ideas on automated judging by computer. If you have 100 pics per photographer per jump, and 4 photographers, you have 400 pics per jump to go through, which is what the Elsinore organizers pretty much said happened. I propose to have a computer vision algorithm to look through the pictures, figure out where each jumper is, which is quite hard, seeing as even I have trouble differentiating the wingsuiters from the backgrond in some pictures. Then apply the judging algorithm and see which picture is the best. If there's anyone out there with programming and/or computer science skills that'd like to see if we can work something out, send me a PM.

In reply to the rest of your lengthy post, I have to say I agree with most of your ideas, although I think the foam noodles are probably not going to work. They're going to be flapping around quite a bit, and catching and holding on to them is going to be a challenge. They would need to stretch a bit too, to allow for movement. One person falling out and holding too long onto the noodle could take out the entire formation... :)
ANyways, good thread... lets keep brainstorming...
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
Quote

When I first read his post I thought he was joking, but reading it again I think I know what he means: figure out how far each wingsuiter was from his intended place in the formation and if the average deviation is less than 2 meters then the formation is considered complete.



I've quickly made 2 examples of what I mean. In the first picture (Pic 17), the average deviation of the line lengths is quite low, the formation looks good. In the second picture (Pic 19), the average deviation is quite high.

Of course, these examples are from 1 picture, which raises the question, if everybody is in their exact slot except for 1 person, the average deviation is still within limts, but the formation is not complete.... what do we do then? Or do we define a max deviation for an individual which is still acceptable?
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
Quote

.. I think the foam noodles are probably not going to work. They're going to be flapping around quite a bit, and catching and holding on to them is going to be a challenge. They would need to stretch a bit too, to allow for movement. One person falling out and holding too long onto the noodle could take out the entire formation... :)

You mean exactly like an RW formation? :P Not to mention CRW, where you can do this to somebody else? :S

They would not need stretch, you just don't fly them at full extension. (They might need to be longer than a foot.) That's why you take foam and not bamboo. You could have them break when stretched - but make one mistake and your grip is gone, you've just taken out the entire formation for the remainder of this jump. :$

I don't know if the flappage would actually be a problem. Maybe for taking grips, but I suspect not for holding it.

RW experience, skills and mindset really, really help in flying this thing I think. I'd like to test it some time, see what it does in the real world.
Johan.
I am. I think.

Share this post


Link to post
Share on other sites
Quote

You mean exactly like an RW formation? :P Not to mention CRW, where you can do this to somebody else? :S

...

RW experience, skills and mindset really, really help in flying this thing I think. I'd like to test it some time, see what it does in the real world.



Okay so you have a point. :$

I guess we can only find out by trying. Where are we going to find foam noodles?

Edit: I have some pics with line lenghts. Blue lines are supposed to be horizontal and black lines diagonal. I'm guesstimating the average length for the black lines is about 1.8 or 1.9 and the blue lines about 1.4 or 1.5.
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
Quote

Not jumping 'cause it's winter??
:D:D:D:D



Well, procrastinating at work actually... But yea when it's below freezing at ground level, it's not worth the effort any more. :P
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
We have LOTS of foam noodles...about .99 at your local walmart in the kids toys/pool area. But I don't think they'll work.
IMO, this is the most productive discussion on the subject yet (trying to find a solution vs criticizing an experiment).
One problem I see with this solution, Costyn, is it only takes 2 axis' into account. So does the grid. How does the Z axis come into play, other than the point being larger/smaller in the image? If someone is in their slot 20' below, they'll still fit the matrix. Should that count?
What about the artistic nuances? How are those quantified? Or are they? If it's all about numbers...

Share this post


Link to post
Share on other sites
Quote

We have LOTS of foam noodles...about .99 at your local walmart in the kids toys/pool area. But I don't think they'll work.



Why do you think they won't work?

Quote

One problem I see with this solution, Costyn, is it only takes 2 axis' into account. So does the grid. How does the Z axis come into play, other than the point being larger/smaller in the image? If someone is in their slot 20' below, they'll still fit the matrix. Should that count?



At the moment, using the grid method, it does count. Eventually I'm hoping there will be a method to reconstruct the position of every skydiver in the sky in a 3d model, as described here. From the 3d model you can now use the z-axis in the calculations. This paper is about reconstructing a static scene, but it's also from 2003. I haven't really looked yet for newer papers, but I hope there are people working on this. It's quite an essential subject for computer/robot vision. There are lots of hits on Google anyways...

But lets work out the 2D stuff first, it's hard enough as it is. ;)

Quote

What about the artistic nuances? How are those quantified? Or are they? If it's all about numbers...



What do you mean by artistic nuances?
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
Yes, I understand that using the grid model, the Z axis somewhat counts, but IMO, it should entirely count. If someone is 24" or more above/below, then that invalidates the formation, or something like that (random number, pulled outta my backside).

I like the 3D concept, and see it as very useful and fairly easy to construct. But, it may be too complex for the Luddites and non-creatives that seem to make up the FAI crowd.

re; artistic nuances;
John Adams' music is mathematically perfect, and is a strain on most ears. Same can be said for many other endeavors. Conversely, the music of Tomita or Nickelback is far from mathematically perfect, but easy on the ears. It's subjective. Does the mathematically perfect formation carry more weight than the asymetrical formation that 'looks' right?'
There is a groove, we've all felt it, where everything fits. It may not be mathematically right but everyone involved knows it is right.
As we get into more vertical formations, some of this becomes easier while some of it becomes more difficult as well.

While formation skydiving does need a consistent means of measurement, should measurebation be the sole validating factor in composing and judging a formation?

Share this post


Link to post
Share on other sites

I don't think they'll work either, at least not for 100-ways. But they might be workable for 4-ways, or highly skilled 9-ways. There's no reason to rule out something that doesn't work for all sizes if it works for some sizes. I say we try it before we dismiss it.

Walmart, Walmart - what's the Euro equivalent of that? We'll figure something out. :)
Because (for now) you only have 2D pictures, you're always going to be limited to 2D criteria. That's not all bad. RW peoples can also cheat with a hand right above a grip instead of holding it, you'll never see it on a photo, it's acceptable collateral IMO.

Another problem with the grid is there is no way to measure 10' exactly, you'll always be scaling to fit the formation. (You will also always be rotating to fit the formation. You might even have to adjust for the angle of a photo if the cameraman wasn't reasonably far above the exact center of the formation.) But spacing has to be the same throughout the formation. Consider it the equivalent of cheating a body position on an RW formation - it has an inherent limitation (RW: you still have to take a grip and your arms are only so long; WRW: you can space it out but it has to be consistent) and it allows for some leeway while staying within the rules.

I have been playing with normalising distances in Costyn's pictures (divide every measured distance by the appropriate average - different for horizontals and diagonals); this means the standard deviation (is the term the same in English as it is in Dutch?) is normalised and comparable between formations as well. I think. Unless Kallend cuts me off at the knees. :)

Johan.
I am. I think.

Share this post


Link to post
Share on other sites
Quote


Another problem with the grid is there is no way to measure 10' exactly, you'll always be scaling to fit the formation. (You will also always be rotating to fit the formation. You might even have to adjust for the angle of a photo if the cameraman wasn't reasonably far above the exact center of the formation.) ..



Don't forget lens distortion in addition to angle of view. Aspherical lenses distort the distances as well.

Share this post


Link to post
Share on other sites
Quote

Yes, I understand that using the grid model, the Z axis somewhat counts, but IMO, it should entirely count. If someone is 24" or more above/below, then that invalidates the formation, or something like that (random number, pulled outta my backside).

I like the 3D concept, and see it as very useful and fairly easy to construct. But, it may be too complex for the Luddites and non-creatives that seem to make up the FAI crowd.



Well, it's actually really hard to construct, but after you have a model in a computer, judging becomes a breeze.

Since we only have 2D data, we can't really take into account the z-axis until we have the 3d reconstruction.

My vision is that eventually we'll have a program you can run on your laptop, tell it the formation you are trying to build, feed it some pictures, and it'll tell you if any of the pictures qualify. That picture can then be further evaluated by officials like Guiness or FAI (gosh, those are scary words to use around here these days. :)
Quote

While formation skydiving does need a consistent means of measurement, should measurebation be the sole validating factor in composing and judging a formation?



Yes. This thread is solely about having an objective way of evaluating a formation for the sake of getting a binary answer: "this formation is complete" or "this formation is not". If not, what's the point of going through all the trouble of measuring the distances?

Johan, standard deviation is the same term in English yes. So, did you get anything interesting out of my numbers? Or do you need more data?
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
If you have mathematical criteria to apply to a 2D map of a formation, I think you can extend them to work on a 3D map. Measure distances in 3D, then start normalising, averaging etc. like you would in 2D.

If a formation is skewed, it may still satisfy some criteria almost as well as when it would not be. Other criteria may not have this property. Skewing the grid may be defined as acceptable, just as rotating and scaling need to be (see above).

I think there can be formation attempts that do not use predefined criteria, but you'll never fly an accepted, official record that way. If that's not what you want anyway, no problem. Sometimes you do, sometimes you don't, I think. A friendly fun 4-way RW counts points until separation instead of in 35 seconds, doesn't it?
Johan.
I am. I think.

Share this post


Link to post
Share on other sites
Quote

Don't forget lens distortion in addition to angle of view. Aspherical lenses distort the distances as well.

As long as we don't forget it, can we for now disregard it? As well as the 3D vs. 2D complication?

For now, I would rather have criteria that work as intended, no false positives, no false negatives, on a slightly simplified model of the formation.
Johan.
I am. I think.

Share this post


Link to post
Share on other sites
OK, the results of the jury are in.

The wing 9-way diamond scores 2.7331 cm horizontal avg and 1.7208 cm diagonal.
The base 9-way diamond scores 1.8429 cm horizontal average and 1.5585 cm diagonal.

I have calculated as if these were two distinct formations; theoretically, they should have the same values for those numbers but they obviously do not. Maybe the infamous aspherical lens distortion? Just angle differences?

Wing diamond normalised horizontal sigma .0416, diagonal .2544, overall .2213
Base diamond normalised horizontal sigma .0737, diagonal .1130, overall .1045

So for these two formations, overall standard deviation of the normalised distances seems to give a usable number. The cut-off value would seem to be around .15 to .20 (or would it still be reasonable to say the wing diamond 'got it' ?) Horizontally, they are not even half bad, but this does not affect the diagonal score really. That's good.

I think this method starts working from 5 flyers; I have no idea yet [how] the cut-off value increases with the number of jumpers.

And what this really needs now is an interactive computer program where you can drag the flyers around and see what the values do in (almost) real time, because I have no idea how the calculated value would be for perfect but skewed formations, crooked ones, indistinct blobs or other abominations. That may be the quickest way actually to determine if this method is valid, robust, accurate and generally correct and useful. Also, testing on some of those formation photos in an earlier thread might not be a bad idea. I'll see if I can find that thread. Do not expect results on that this year ..

A possibility would be to split diagonal distances into / and \ for normalisation. You could calculate away quite some skew that way, I think. If all you have is a picture taken at an angle anyway, you might want to do that.
Johan.
I am. I think.

Share this post


Link to post
Share on other sites
Quote

I have calculated as if these were two distinct formations; theoretically, they should have the same values for those numbers but they obviously do not. Maybe the infamous aspherical lens distortion? Just angle differences?



The wing is quite a mess compared to the base, so we'd expect the values to be a lot higher.

Quote

(or would it still be reasonable to say the wing diamond 'got it' ?) Horizontally, they are not even half bad, but this does not affect the diagonal score really. That's good.



In my book, the whole formation should look like the base, so no, the wing didn't get it. The left wing was problematic throughout the event for unknown reasons.

Quote

I think this method starts working from 5 flyers; I have no idea yet the cut-off value increases with the number of jumpers.

And what this really needs now is an interactive computer program where you can drag the flyers around and see what the values do in (almost) real time, because I have no idea how the calculated value would be for perfect but skewed formations, crooked ones, indistinct blobs or other abominations. That may be the quickest way actually to determine if this method is valid, robust, accurate and generally correct and useful. Also, testing on some of those formation photos in an earlier thread might not be a bad idea. I'll see if I can find that thread. Do not expect results on that this year ..



I thought the exact same thing, and I've been doing some googling today, looking for a Java app that allows you to place points (nodes) connect them with lines (edges) and move the points around, I've only found some libraries so far. I'd code it up myself, but don't want to re-invent the wheel, and my Java skills are quite rusty unfortunately. If anyone else finds anything, please post it here! Also, any Java programmers around? Here's looking at you, Vidiot/Klaus. :)
Quote

A possibility would be to split diagonal distances into / and \ for normalisation. You could calculate away quite some skew that way, I think. If all you have is a picture taken at an angle anyway, you might want to do that.



A picture taken from a different angle (to the side or from behind or in front) should not affect the diagonals should it? They will become steeper (from the sides) or wider (from front/behind) but relatively they should stay the same. Also lens distortion should only minimally affect the numbers, but we'd need the app to test it out.

Cheers

Cheers
Costyn van Dongen - http://www.flylikebrick.com/ - World Wide Wingsuit News

Share this post


Link to post
Share on other sites
Quote


IMO, this is the most productive discussion on the subject yet



I was just thinking the same thing :)
I have been playing with simpler solutions which do not require measurements, rather looking for straight lines across the formation. Varying the thickness of the lines for tolerance.

As pointed out previously the tricky part is finding a balance for the position tolerance of each flyer. 0.0 Perfection vs random mess in the sky.

IMO the formation as per the dive plan should be visible at first glance, think center fold photo :)
All of my ideas so far suffered the "Yuri scatter" flaw [:/]

There isn't a world record for the largest formation of planes - perhaps for the reasons we are discovering right now!
BASEstore.it

Share this post


Link to post
Share on other sites
Quote

Quote


Another problem with the grid is there is no way to measure 10' exactly, you'll always be scaling to fit the formation. (You will also always be rotating to fit the formation. You might even have to adjust for the angle of a photo if the cameraman wasn't reasonably far above the exact center of the formation.) ..



Don't forget lens distortion in addition to angle of view. Aspherical lenses distort the distances as well.



As long as you know the distortion, you can correct for it.

BTW my research is in crystallography, where we use X-ray diffraction pictures to figure out where the atoms are in 3-d. The mathematics is all well known and (relatively) straightforward. I wrote thousands of lines of code in my time to do such analyses, much of which became standard in labs all around the world.

This strikes me as a very similar problem, easier in some respects (X-ray diffraction is not as "linear" as optics) and more difficult in others (knowing exactly where the camera is located relative to the formation), but quite do-able.
...

The only sure way to survive a canopy collision is not to have one.

Share this post


Link to post
Share on other sites
Quote

Calling John Kallend!

Discuss ..



OK, there were two points to my suggestion, which is, I guess semi-serious as it could be deone but would take quite a bit of work.

One is to allow distortions from the ideal formation. Let's say the design grid for a diamond has a 75 degree angle at the apex. Does it really matter if the skydivers fly it with a 78 degree angle? I suggest not. So I'd allow a "homogeneous deviatoric" distortion of the grid to find the best fit to what the skydivers did (in other words, allow a change in the overall angle of the diamond, but still keep every cell the same shape). Secondly, judging 3m spacing (or whatever) is difficult - a 0.1m error in each skydiver's position adds up to several m error between base and outside on a large formation. So I'd allow a uniform (homogeneous) size distortion too. All cells are the same size, but it might be slightly bigger or smaller than the design size. After all, even RW formations are judged "complete" if they breathe in and out by several meters as long as everyone is in the "right" place).

Finally, having a uniform grid that best fits the formation, calculate the rms error between the jumper's positions and the centers of each cell.

Edited after reading the above goobledegook.

Imagine drawing your ideal formation on a sheet of rubber. You then allow the rubber sheet to be uniformly stretched or distorted until the actual formation best corresponds to it . Each cell is still the same shape and size as all the other cells, but may be slightly changed in shape and size from the original design. Then you find the average error between where the skydivers are and the centers of the cells.
...

The only sure way to survive a canopy collision is not to have one.

Share this post


Link to post
Share on other sites
install the tube shape extensions on the arm and legs..
Correct formation is the one were the tube overlaps/coves the another jumper arm or leg or head.
extension can be i.e. 30 cm long meaning that distance between jumpers can be no bigger than 29 cm

one of the most important part of recognizing ''the thing'' is simplicity. in a second / minute it should be clear if the record is there or not. Grips are the ''rule'' in skydiving records... IMHO to create or have the rule which depends on lens distortion, calculator, ruler etc... is no good. it will not be excepted in long term
do something similar to it and job is done...

happy 2009.
Robert Pecnik
[email protected]
www.phoenix-fly.com

Share this post


Link to post
Share on other sites
Quote

There isn't a world record for the largest formation of planes - perhaps for the reasons we are discovering right now!

You mean there's a record up for grabs? Would four Otters count? :P They were in the same frame on, oh, any number of photographs. :)
Johan.
I am. I think.

Share this post


Link to post
Share on other sites
Agree that the complex rules are probably not the best way. These tubes are they soft? If so won't they blow back making the distance they represent between two jumpers depend on glide angle? Still some length based on formation size and observed glide can be predetermined.

Don't drop your tube :)

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

0