Video Metrics - Vid Analysis Data Points for PV Performances

This is a forum to discuss advanced pole vaulting techniques. If you are in high school you should probably not be posting or replying to topics here, but do read and learn.
User avatar
KirkB
PV Rock Star
Posts: 3550
Joined: Mon May 19, 2008 6:05 pm
Expertise: Former College Vaulter; Former Elite Vaulter; Former Coach; Fan
Lifetime Best: 5.34
Favorite Vaulter: Thiago da Silva
Location: Vancouver, BC, Canada

Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby KirkB » Thu Mar 12, 2009 9:21 pm

In the "Push off" thread, we're talking about how to determine vaulter "efficiency". We have discussed "vaulter height", "pushoff", "grip", "takeoff speed", and "pole selection" (pole flex) as being factors affecting the efficiency of a vaulter clearing a certain crossbar height. Our discussion has centered around the 6.00m club members, primarily because (a) they're elite vaulters that we aspire to emulate; and (b) their data is readily available.

Linthorne provides a fairly thorough mathematical anaysis on the factors and formulae that cover this topic. The main issue I have with his paper is that he arbitrarily adds 1.00m to his calculations to cover the "average pushoff" of an elite vaulter. Clearly, the pushoff varies significantly from one vaulter to another. And clearly, the EFFICIENCY of each vaulter is hidden somewhere in their pushoff.

One objective that I propose of this video analysis methodology is to try to nail down additional factors that can "rationalize" or "explain" why one elite vaulter's pushoff might be more or less than another's. I'm quite certain that the speed of the DOWNSWING ... once measured ... will become a VERY important consideration in determining a vaulter's pushoff or efficiency. But there's others.

Another objective is to be able to scale this video analysis back down to your own vault, or to the vaulters that you're coaching. My hypothesis is that we will find a continuum of metrics from beginners thru intermediates and elites ... male and female. If you're "off the scale", then you're either doing something terribly wrong ... or terribly right! :)

A further objective could be to scale it down to elite women vaulters, to try to identify (more precisely) in which part of their vault women are relatively slower or weaker then men.

Another is to take videos of your subject vaulters at various times during a season ... or from season to season ... and compare their progress over time ... using these common data points.

Yet another idea is to standardize the metrics captured from vids, so that we can communicate with a common language ... including common data points taken from vids at key points.

I'm sure there will be other good things that will materialize out of this idea.

Some of this work is already done. I'll have to dig back on my notes from 4 years ago, but that's when I was helping Pat Licari analyze vids of Brad Walker, and I manually took some of these specific data points during his vault and attempted to compare them to other elites. That idea really never got off the drawing board and into "production" but today ... it's better late than never.

In communicating with DJ about that idea, I heard from him that he had done a lot of earlier work dating back to the 1980s with Mike Tully, timing and measuring his vault parts. He has mentioned many times on PVP the importance of the timing of each vault part, and the fact that an elite vaulter takes about the same amount of time "in the air" as a college or HS vaulter.

I owe a lot of credit to DJ for helping me to formulate my ideas about these Video Metrics.

He did much of the original work, and I followed his footsteps. I haven't talked to him about this yet, but I'm also hopeful that he will contribute to this thread in a significant way ... his time and interest permitting.

The idea is actually very simple ...

1. Videotape the vaulter, or use readily available vids of vaulters found on the internet.
2. Take video frame samples of a vault.
3. Take data point samples (by both time and distance) of the vault at key body positions.
4. Compare these data points to other vaults.

The points that I propose (which we can discuss and decide upon where my ideas are suboptimal) are:

1. The frame just before the takeoff foot leaves the ground (compared to the frame just before that).
2. The frame just before the pole hits the box (compared to the frame just before that).
3. The frame just before the "I" (Whip) position at the end of the downswing (compared to the frame just before that).
4. The frame closest to the "flat back" position (compared to the frame just before that).
5. The frame just before the top hand releases the pole (compared to the frame just before that).
6. The frame closest to max hip height (compared to the frame just before the top hand releases the pole).

Why these 6? Well, (1) gives you the runway speed; (2) gives you the "jumping impulse" speed; (3) gives you the "downswing" (max trail leg swing) speed; (4) gives you the "upswing" speed; (5) gives you the extension speed; and (6) gives you the speed of vertical deceleration from pole release to bar clearance. Unlike the others, this deceleration speed isn't anything that you can do anything about ... you're at the mercy of gravity then. It's simply a RESULTANT metric, caused by the other metrics measured by 1-5.

So what do we want to measure on these sample frames?

Speed = Distance / Time

If we can capture time and distance, we can determine speed.

And if we know the speed of each of these 5 key vault parts, we can compare them, analyze them, make decisions on if they're too fast or too slow, and ultimately use them as a guide for focussed training to improve specific vault parts.

To say this another way, we already know that the faster we're running just before takeoff, the higher we can vault (all else being equal). This idea takes it a step further, to see which vault parts AFTER the run are too slow.

Once we identify them, we can fix them! If you don't know what's broken, you can't fix it!

The TIME is computed from the camera shutter speed.

The DISTANCE is measured by laying the target frame over the frame just before it. The distance between the "key body point" on the 2 frames, multiplied by the scaling factor of the frame, gives the actual distance in meters and fractions of meters.

What are the "key body points"? They vary per frame. You want to pick body points that are easy to identify, and represent what you're measuring.

The CoM is an important body point, but it will vary according to your various body positions during a vault. A good approximation of the CoM is the point at which a vaulter's shorts intersect with his jersey, just below his belly button (as opposed to his back side). Especially with contrasting colored shorts and jersey, this is usually an easy point to identify.

A better idea ... if you think of it before videotaping ... is to mark a dot (visible on vid frames) on the side of a vaulter's shorts or jersey in a contrasting color. While you're at it, you could mark dots on certain fulcrum points of the body, such as the shoulder, hip, knee, and ankle ... but that's more than we need for my basic idea of capturing common data points (so pardon the digression :)).

For (1) and (2), take the shorts/jersey intersect point; for (3) take the intersect where the trail leg ankle intersects the heel of the track shoe; and for (4) (5) and (6) take the shorts/jersey position.

So now we have times and distances. Now we need to SCALE the distances. Here's how ...

This is Pogo's post today on the "Push off" thread, that partially inspired me to get this post started (although I've been planning to do this for a few days now) ...
Pogo Stick wrote: I think grip data are more accurate than any other data (except bar clearance :)).
2" error is 1% of pole length (very accurate) or 5% of push off value (acceptable). ...

If you know a vaulters standing height ... which is usually well publicized ... and you have vid of the vault ... you can determine his grip ...

1. Find the frame where he's finishing his downswing, and his trail leg is aligned with his torso and top arm (the "I" position). Alternately, you can use the frame just before takeoff, as long as his takeoff leg is directly under his torso.
2. Measure his height on that frame ... from heel to top of head. Adjust for any abnormal head or ankle positions (if any).
3. Find a frame during his run or plant where the entire pole is visible ... and straight.
4. Measure his pole length, scaling by his known height and his height in the vid frame. Since poles are manufactured in common lengths, you can verify the accuracies of your 2 measurements thus far, and adjust accordingly (assuming his pole isn't cut to a custom, unknown length).
5. Using the same (verified) scale, measure the distance from the top of this known pole length to his top hand, and subtract that from the pole length. That's his grip!

There will be some inaccuracies in this method. If you simply measure directly on a computer screen, your "pencil point" will be quite large relative to the distances you're measuring. If you don't have any high-tech measuring devices to measure the images directly, you can mitigate this by printing the frame(s), and then enlarging them on a photocopier. All data points will be enlarged to the same proportions, so using low-tech measuring tools (pencil, ruler) on the enlargements will be more accurate.

Since you're wearing the engineering hat Pogo (and have lots of time on your hands ;)), I leave it to you to determine how accurate this measuring method is.

OK, now we have time, distance, and scale!

So now we can measure REAL distances from vaulting vids! ... and get REAL speeds of vault parts to compare ... and figure out what factors (speed, distance, timing) of vault parts AFTER TAKEOFF affect a vaulter's pushoff. THEN, we can truly measure which vaulters ... short, tall, light, heavy ... have advantages over each other in specific vault parts ... with the ultimate objective of determining which techniques we should emulate, how we should emulate them, and why.

Along the way, we should be able to mathematically compare disparate models, such as Joe Dial's "drive vault model, to Bubka's "Petrov model".

One last question ... should we compare SPEEDS, or DISTANCES?

We could do either, but distances will need to be scaled quite a bit if we're comparing a HS vaulter to an elite ... and you can't just compare the distance between 2 frames - you need to compare distances between one vault part and the next (separated by frames 1-5). Since DJ has already determined that the time "in the air" is similar for ALL levels of vaulters, comparing SPEEDS might be more straight-forward. Really, just comparing SPEEDS keeps things very, very simple!

You can also compare the SPEED DELTAS between one vault part and the next (by counting frames). This gives you the ACCELERATION or DECELERATION of the body during that vault part. Pogo can give us the formula for that! But to do that, we need one more metric - the speed of the CoM (approximated by the shorts/jersey intersect) at the end of the downswing (Frame #3). We can't compare trail leg swing speed to "torso" speed ... that's comparing apples to oranges.

Will this idea work? I dunno! We'll have to put it into practice! So far this is just a theory! :D

Pogo and DJ ... can you help me with this? Carolina? Tim? Anyone else interested in this?

Kirk
Run. Plant. Jump. Stretch. Whip. Extend. Fly. Clear. There is no tuck! THERE IS NO DELAY!

dj
PV Enthusiast
Posts: 1858
Joined: Fri Oct 08, 2004 9:07 am
Expertise: Coach
Contact:

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby dj » Fri Mar 13, 2009 1:10 pm

hello..

1. The frame just before the takeoff foot leaves the ground (compared to the frame just before that).
2. The frame just before the pole hits the box (compared to the frame just before that).
3. The frame just before the "I" (Whip) position at the end of the downswing (compared to the frame just before that).
4. The frame closest to the "flat back" position (compared to the frame just before that).
5. The frame just before the top hand releases the pole (compared to the frame just before that).
6. The frame closest to max hip height (compared to the frame just before the top hand releases the pole).


i had some reservation about number one… simply because of the various plant position.. out.. under etc.… by catching the/a point where the knees are even, thighs parallel at the plant… I felt the COM would be near the same point (body vertical hopefully) and therefore travel the same distance base on grip height. If you wait until the toe is just about to leave the surface the under vaulters or an under vault would have the COM in advance of an out vault.. which would not compare well for the distance travel in X time.. now that may be some kind of “odd” thinking.. I mean the under vaulter could get “yanked” onto there back fast.. but then “stall” so the numbers would be as relative as the total distance the center of mass travels from a vertical position, because the “in” side of the body at vertical is what, to me, will give us the most consistent, comparative measure.. maybe get both and let the reader see what is relative.

Number 2 is very hard to measure… don’t know if I mentioned it before but at the pit in Northridge I put a “lighting” system to see what was going on at the takeoff and when the swing was occurring on the best vaults.. I used this with Tully and fourteen other vaulters and decathletes I was working with at the time….. What I did was put a “sound” switch, clap on clap off type that would come on when the pole hit the box.. to this switch I put a yellow light attached to a hurdle in the camera sight between the camera and the vault.. I set the camera 90 degrees to the “middle’ of a vault.. I put a switch across the top back of the box.. long doorbell type so I could see where the vaulter were in the swing when the pole hit the switch… with the thought that when the pole hits at that point it’s penetration may stop and the pole will slide to the corner of the box.. my thoughts also told me the vaulter had better be past the “back flat” to the runway position and swinging fast to get back or they would be thrown forward into the bar or pit…

I think we can collect some great important data from a lot of vaults without having a “standard” camera position.. I have always tried to think that for every action there is an equal and opposite reaction… and the time and distance of the rotation around the top hand grip has to be important in the big scheme of things…

changes in rotational speed about an axis is called torque.
Torque is the tendency of a force to rotate an object about an axis………

dj

User avatar
KirkB
PV Rock Star
Posts: 3550
Joined: Mon May 19, 2008 6:05 pm
Expertise: Former College Vaulter; Former Elite Vaulter; Former Coach; Fan
Lifetime Best: 5.34
Favorite Vaulter: Thiago da Silva
Location: Vancouver, BC, Canada

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby KirkB » Sat Mar 14, 2009 12:19 am

dj wrote: i had some reservation about number one… simply because of the various plant position.. out.. under etc.… by catching the/a point where the knees are even, thighs parallel at the plant… I felt the COM would be near the same point (body vertical hopefully) and therefore travel the same distance base on grip height. If you wait until the toe is just about to leave the surface the under vaulters or an under vault would have the COM in advance of an out vault.. which would not compare well for the distance travel in X time..

You've convinced me on this one. All we're really trying to do with #1 is measure horizontal takeoff speed ... there's no need to catch any takeoff impulse yet. And "thighs together" is a very consistent data point to catch for all vaults and all vaulters. :yes:

dj wrote: Number 2 is very hard to measure…

Yes, I was worried about that too, because of the slow shutter speed on most consumer vidcams, combined with the fact that the jump impulse is a very short duration. So I sampled a few youtube vaults ... this is where the theoretical meets the practical (you've already figured this out) ... as suspected ... for #2 ... it's the luck of the draw whether you get the vaulter in exactly 2 frames ... the first one being "thighs together" (as you say), and the next one being Frame #2 - just before the pole hits the box. You usually get one or the other, but not both - it's just too fast for a slow-speed (actually normal speed) camera.

The second issue is when capturing data points of "drive vaulters". I looked at Jack Whitt's takeoff frame-by-frame, and in his case, there IS NO JUMP IMPULSE. This is by design, as Tim has explained to us many times. No problem ... it's just a different model ... so there's no jump impulse to measure. For drive vaulters, you have to take the "thighs together" frame as the last one (to measure horizontal takeoff speed). Any later than that, and the CoM slows down, due to loading the pole before takeoff.

I'm thinking that for Petrovers, #1 would be with "thighs together" (and the frame before that), and #2 would be as close to #2 (as originally stated) as possible. For drive vaulters, #1 would be as per Petrovers, and #2 would simply be duplicated from the same 2 frames. Every subsequent frame sample should be identical between the 2 models. At first glance, it might appear that this biases the speeds according to model. But if you consider #2 as the start of the bend rather than the end of the takeoff, it makes sense, and we're back to apples and apples. :idea:

It's not going to be perfect no matter how we slice it. Can you think of another way to resolve this?

dj wrote: ... I put a “lighting” system to see what was going on at the takeoff and when the swing was occurring on the best vaults .. I ... put a “sound” switch, clap on clap off type that would come on when the pole hit the box.. to this switch I put a yellow light attached to a hurdle in the camera sight between the camera and the vault.. ... I put a switch across the top back of the box.. long doorbell type so I could see where the vaulter were in the swing when the pole hit the switch…

Wow! That was quite high-tech for the late 1970s / early 1980s! ;)

Kirk
Run. Plant. Jump. Stretch. Whip. Extend. Fly. Clear. There is no tuck! THERE IS NO DELAY!

User avatar
KirkB
PV Rock Star
Posts: 3550
Joined: Mon May 19, 2008 6:05 pm
Expertise: Former College Vaulter; Former Elite Vaulter; Former Coach; Fan
Lifetime Best: 5.34
Favorite Vaulter: Thiago da Silva
Location: Vancouver, BC, Canada

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby KirkB » Sat Mar 14, 2009 12:34 am

In addition to capturing data points of the VAULTER to compute his speed at key points of his vault, we can also capture data points of the POLE, to compute its speed at these same key points.

Points of interest are:

1. Speed of pole as it rotates towards vertical.
2. Speed of compression of the pole as it coils.
3. Speed of recoil.

All these data points are in relation to the grip of the top hand on the pole.

The chord of the pole (ever changing) rotates towards vertical in an arc. Between any 2 adjacent frames, a straight line is a fair approximation. On each frame that's used to measure the CoM, the "grip point" can be captured as well. We can then see where the pole speed increases or decreases (in the horizontal direction) towards "grip release" (aka "pole release" or "vertical").

The compression/decompression of the pole is measured by the distance between the grip and the butt of the pole butt in the box. The time taken between frames ... at each key data point ... will reflect the speed of this pole action. Once we know the pole speed, we can compute its acceleration/decelaration between certain key points in the vault.

Remember that the pole is doing its thing whilst the vaulter is doing his thing. So with both happening simultaneously, the data points are capturing the NET SPEED of the vaulter/pole system. There's all kinds of interesting formulae that can be used in modeling the transformation of the vaulter's kinetic energy to/from the pole's potential energy.

These formulae are a joy for someone like Pogo ;), but a nightmare for the rest of us ... non-engineers.

I'm not trying to figure out any of those details here ... I'm just trying to establish "What are the correct data points to capture?" so that the formulae can be applied once we capture some data.

If we're able to capture this for ... say the 6.00m club members ... then Pogo can work his magic from there! ;)

Kirk
Run. Plant. Jump. Stretch. Whip. Extend. Fly. Clear. There is no tuck! THERE IS NO DELAY!

baggettpv
PV Master
Posts: 707
Joined: Mon Sep 09, 2002 2:04 pm
Location: Oregon City, Or
Contact:

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby baggettpv » Sat Mar 14, 2009 1:32 am

The top hand movement is a measurement of the takeoff angle. Put a clear plastic overhead sheet on the tv and plot this point frame by frame. Simultanously plot the shoulders, hips and trail leg foot. When they go thru a 45 degree angle from the box they should be a straight line no matter how high they grip (1984). Nice stats KB, are we trying to explain or develop a plan?

Rick Baggett
WSTC
Good coaching is good teaching.

baggettpv
PV Master
Posts: 707
Joined: Mon Sep 09, 2002 2:04 pm
Location: Oregon City, Or
Contact:

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby baggettpv » Sat Mar 14, 2009 1:34 am

And by the way, what is the plan? Simply.

Rick Baggett
Good coaching is good teaching.

dj
PV Enthusiast
Posts: 1858
Joined: Fri Oct 08, 2004 9:07 am
Expertise: Coach
Contact:

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby dj » Sat Mar 14, 2009 1:40 am

good morning

the lighting system gave me great feed back... for one.. it gave me the take off position as to "when" the tip made contact.. i was trying to eliminate the "negative" bend (if you drop the pole wrong the pole bends down before the grip is all the way up)Tully was getting on the plant.. I was looking for the pole to slide/glide in as the grip went from the ear to full extention .. and then the pole would hit the back of the box.. and the toe would leave the ground..

I called the action on the last two steps.. "turning the pole over".. David Butler calls it a "filp-pop"

when Tully did it right he felt 3 to 4 inches out.. he and i would leave the camera running and put verbal "notes" on each vault so when we sat down to watch.. we had the "MID" the "feel" from him.. and what i saw.. those were done with the old super 8 camera...

that's where i found the "back flat" position was important... that taking off out and not “force’ bending the pole was the way to go.. (and Charlie on here will remember we had some pretty good “chats” about taking off a foot under!!!)..

we might get some pretty good data from this process..

Ladyvolscoach has some clips of T-Mack and Daniel both jumping 5.90.. he over laid them with a great result. maybe we can add those to this thread, both separatly and overlayed and matched up at the plant...

Later

Time for morning work out…

dj
Last edited by dj on Sat Mar 14, 2009 12:38 pm, edited 1 time in total.

User avatar
KirkB
PV Rock Star
Posts: 3550
Joined: Mon May 19, 2008 6:05 pm
Expertise: Former College Vaulter; Former Elite Vaulter; Former Coach; Fan
Lifetime Best: 5.34
Favorite Vaulter: Thiago da Silva
Location: Vancouver, BC, Canada

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby KirkB » Sat Mar 14, 2009 2:30 am

baggettpv wrote: Nice stats KB, are we trying to explain or develop a plan?


baggettpv wrote: And by the way, what is the plan? Simply.

I'm trying to find (and perhaps agree upon, or at least reach some consensus on) some commonality in how we analyze vids. Not by distance or body positions and angles, but by time ... each frame being a consistent time duration. DJ has already done some work in this area.

I think we can still subjectlvely analyze body positions frame-by-frame, but this time analysis would be more objective ... more mathematical.

I'm thinking that we can turn these data points into formulae that will allow us to [eventually] extrapolate to determine "vault efficiency" ... whatever that means. See: "Push off" thread here http://polevaultpower.com/forum/viewtopic.php?f=38&t=17039 for some discussion around this idea ... which continues to evade us. :dazed:

This "plan" is still on the drawing board. What you see here thus far is what you get! I'm hoping we can develop this as a PVP collaboration. Your contributions are very welcome! :yes:

I've just recently downloaded Firefox DownloadHelper (http://www.downloadhelper.net/), so I can now (finally) analyze youtube vids frame-by-frame. The quality's not there, but youtube vids are so ubiquitous that I'm looking for a "standardized" way of comparing one athlete's vids to the next, or to analyze improvement of a single athlete thru the course of a meet, a season, or a career.

It's all about establishing some "Video Metrics", so that we can discover strengths and weaknesses in vaults ... mathematically ... by way of the TIME dimension.

One challenge is the poor vid quality and slow shutter speed of most consumer vidcams, but we have to allow for this ... and not require high speed vidcams for this to work. i.e. It has to be PRACTICAL. :yes:

I need someone with good knowledge of vidcam shutter speeds to chime in here. :star:

Kirk
Run. Plant. Jump. Stretch. Whip. Extend. Fly. Clear. There is no tuck! THERE IS NO DELAY!

dj
PV Enthusiast
Posts: 1858
Joined: Fri Oct 08, 2004 9:07 am
Expertise: Coach
Contact:

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby dj » Sat Mar 14, 2009 4:11 am

good morning again

unless you are 90 degrees to the vaulter.. the angle distorts the tracking of the top hand grip..

frame counting is a very good method of determining the time… the first trajectory I got on Bubka, I think 1985, had the frames from takeoff to release.. even though I didn’t know the film speed I could count how many frames it took to maximum bend and compare it to the various position by percentage.. once I knew the total time I did the math and got numbers for takeoff to maximum bend… maximum bend to release.. etc..

that is why I timed T-Mack.. his total Vault was 1.43-1.44 seconds and it took him .49-.51 from takeoff to maximum bend.. when he went to a bigger pole on the next vault, which had the same run and same speed and same takeoff.. it took him longer to get maximum bend and the takeoff angle was greater… these are the reasons I am a strong believer that “stronger” is not always better (it is actually the wrong thinking for a vaulter) and that the trajectory of the top grip rick is talking about is created by the forces applied at the plant/takeoff and the stiffness of the pole… period. Not from a jump or intent to jump.. the jump can only help in having a higher COM at the takeoff.. and will only help if you don’t lose speed when you do it..

Colwick has no “jump” or even an “impulse” .......but he can vault high because he carries his speed right onto the pole…….

Later

dj

User avatar
Pogo Stick
PV Pro
Posts: 427
Joined: Tue Sep 25, 2007 11:42 pm
Expertise: Former "College" Vaulter, Masters Vaulter
Lifetime Best: 4.70/15'5
Favorite Vaulter: Władysław Kozakiewicz
Location: Vancouver, Canada; Split, Croatia
Contact:

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby Pogo Stick » Sun Mar 15, 2009 5:46 am

KirkB wrote:Since you're wearing the engineering hat Pogo (and have lots of time on your hands ;)), I leave it to you to determine how accurate this measuring method is.

OK, now we have time, distance, and scale!

So now we can measure REAL distances from vaulting vids! ... and get REAL speeds of vault parts to compare ... and figure out what factors (speed, distance, timing) of vault parts AFTER TAKEOFF affect a vaulter's pushoff. THEN, we can truly measure which vaulters ... short, tall, light, heavy ... have advantages over each other in specific vault parts ... with the ultimate objective of determining which techniques we should emulate, how we should emulate them, and why.
Kirk


Disclaimer first: I am holding B. Sc. of EE, but I never worked as engineer. Here in Canada my diploma is not even worth the paper it's printed on, but this is another story.
My first thought was why not use some existing software like Dartfish? I have no idea what can be calculated and measured with Dartfish but ladyvolspvcoach is using it and she should be the best to write review.

Let see how useful some hardware and technologies can be for video analytic.

Camcorders:
I made few calculations based on my current camcorder Canon ZR950 (MiniDV, $220 at Future Shop or Best Buy for you in US). Canon marketing advertise it as 1.07 Megapixel, entry-level camcorder. The truth is the camcorder has 550,000 pixels in 16:9 mode without using Advanced Zoom (don't ask me what is this - engineers don't read manuals). Translated to your monitor resolution this is 980x556 pixels. If horizontal view range is 10m (33 ft) wide then each pixel "covers" 1 cm. That means error is +/- 1 pixel or +/- 1 cm or 0.4" or 0.1% of visible area. This is very good.
10m horizontal view fields is enough to cover standards, take off position and 4-5 meters before (last two steps). Vertical view range will be 5.6m (18ft) for 16:9 format and 7.5m (25ft) for 4:3 format.
If camera shooter speed is not high enough the result will be blurred picture. For 1/1000 shooter speed and 10m/s velocity blurred area is around 1cm or 0.1% error.
Another source of inaccuracy is FPS rate (Frames Per Second). FPS is defined by standard and NTSC has 30 frames per second or 60 half frames (don't even think to ask what's this - it is internal engineering stuff;)). That means camcorder is recording one picture every 3.3 hundreds of second (milliseconds or ms). For running speed of 10m/s that means you are missing 33 cm (more than 1 feet) of running between two consecutive pictures. This is equal to 3.3% error. Good news: for lower speeds (push off phase, for example) the error decrease proportionally.

TV:
North American TV standard (NTSC) has 525 horizontal lines and 480 vertical lines, but in real life you will never get full number of lines from video sources. According to independent tests ZR950 resolution is around 300 horizontal lines and 320 vertical lines. You will not find this info on Canon website, at least not at same page with "1.07 Megapixel" claim. 10 m area divided by 300 lines means each line "covers" 3.3cm (1.3"). That's 0.33% error. For comparison most TV stations broadcasts around 300 horizontal lines, typical VCR has 200-250 and typical DVD player around 400 horizontal lines.

Other:
1. Camcorder optics, especially on cheaper models can be pretty low quality. All kind of errors like image distortion, blur, optical aberration, etc. are the problem for accurate measuring of height, length or distance. I cannot provide exact numbers for that.
2. For proper distance measuring and calculating scaling factor you need reference point in frame - some object with known dimension like 1 yard long stick or markers on runaway. Camcorder position is also important. If you are not perpendicular to marker you need to adjust (simple trigonometry) but you need to know exact distance from reference point. Bar height can be good reference point for calibration, of course, if is visible.
3. Video recorded without tripod and/or panning just add more blurring and inaccuracy.

All together, for someone running 10m/s, an error or inaccuracy causing by either missed or blurred parts of run can be 6-10%. If this is too high error margin for you, Engineering Marketing Department together with Statistics Marketing Department invented something called average error, interpolation and approximation. If, for example, the exact frame of take off (when foot leave the ground) is missing, we can still made pretty good estimate about time and position of take off using frames before and after that moment. We can say that an average error will be half of that value. That means 3-5% or 10-15 cm. As I said in another post, any engineer will kill for such low error.

Video formats and compression:
Compression kills details and blur the picture. Final video quality depends on video source quality, video codecs, output resolution and compression rate. Some video formats are saving video as series of independent pictures (that's good) and others are saving only difference between 2 or more frames (that's bad). Think QuickTime vs. YouTube. YouTube videos are infamous by low quality, but recently they made some changes and there are more and more High Definition Videos with much better quality.
For illustration here is a detail takenfrom the same miniDV tape, saved in 4 different formats and enlarged 400%:
1. QuickTime (.mov)
Image

2. High Definition Video (.avi, 720p (progressive scan), MPEG-2)
Image

3. High Definition Video (.avi, 720i (interlaced), MPEG-2)
Image

4. YouTube Flash Video (.flv), downloaded and converted back to QuickTime format
Image

First two are very good with pretty sharp with well-defined edges. Last two are almost useless.

To summarize:
- You need high quality, high resolution video, recorded with the highest possible shooter speed. Using tripod is a must
- For best results, watch the video directly from camera on monitor or good HDTV without any conversion
- I have no exact numbers, but it looks that measuring distance will be pretty inaccurate, especially on low quality videos. Without reference object (something with known dimensions) calculating scaling factor can be very inaccurate
- Don't use YouTube videos except maybe high quality ones
- Listen dj's advice and calculate time for each phase. It is like having stopwatch with accuracy of 3.3 hundreds of second (or 1.7 ms if you are using interpolation)
-- Pogo

"It is not necessary to change. Survival is not mandatory." W. Edwards Deming

User avatar
KirkB
PV Rock Star
Posts: 3550
Joined: Mon May 19, 2008 6:05 pm
Expertise: Former College Vaulter; Former Elite Vaulter; Former Coach; Fan
Lifetime Best: 5.34
Favorite Vaulter: Thiago da Silva
Location: Vancouver, BC, Canada

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby KirkB » Sun Mar 15, 2009 3:57 pm

Pogo, thanks for all that!

Pogo Stick wrote:My first thought was why not use some existing software like Dartfish? I have no idea what can be calculated and measured with Dartfish but ladyvolspvcoach is using it and she should be the best to write review.

Dartfish is probably a great tool ... but it's out of the price range for what I'm hoping we can do here ...

When Rick asked me ...
baggettpv wrote:And by the way, what is the plan? Simply.

... I'm afraid that I never gave him (or the rest of you) a simple enough reply.

In my business (computer software), if you're developing a new product and are trying to convince stakeholders of its worthiness, you need to make your "sales pitch" so SIMPLE and so SHORT that if you happen to get on an elevator with a prospective product sponsor or investor, you can explain your idea to him ... and SELL it to him ... so that he UNDERSTANDS it ... enough to be willing to listen to more about it later. i.e. You get "the nod".

So here's my Elevator Pitch ...
Pole Vault Video Metrics is a process of analyzing amateur PV videos - to help aspiring young vaulters compare the speeds of different parts of their vault, and to compare their speeds against elite vaulters. This will help them improve their technique, strength, speed, and ultimately their PR - by spotting their weaknesses, making improvements thru focussed training, and tracking their progress.

The process is so simple that many vaulters and coaches will be able to do this themselves, without the aid of expensive software or hardware tools, and without the help of experts. All they need to do is to publish their vids on YouTube, follow the Video Metrics process, and then compare the speeds of their vault parts to their previous speeds - and to elite vaulters' speeds. Other than their video camera and their PC, there's nothing else to buy - just follow the Video Metrics process!

OK, that's quite a mouthful, but I think it covers it! I think it can be explained during an elevator ride ... as long as it's a tall building ... and you're both going to the top! :yes: :D

Pogo, you can see that some of what I just said is fundamentally the opposite of what you say is the MINIMUM of what we need for good data. You say we need higher quality cameras ... I say we make do with what we already have. You say we need Dartfish ... I say that kills the idea, because it won't be available to the masses of do-it-yourself coaches and vaulters. You say there's certain error rates that might be insurmountable - to the point of giving us dodgy data ... I say "we have to find a way", and not let that stop us!

Yes, this is an ambitious project. Some might even think it's fool-hardy. I think not. I think it can be done. We just have to find a way!

My optimism is partly due to the fact that today, we can subjectively view poor-quality amateur vids, and we're still able to provide "reasonable" subjective analysis of the vaulter's technique. Yes, it's crude, but it's FAR, FAR better than the alternative of a vaulter describing in WORDS what his problem is, and us advising him based on that! Vids are GOOD ... they're better than nothing ... no matter how poor the quality!

So if vids help us to SUBJECTIVELY analyze a vaulter's technique, then why can't we OBJECTIVELY analyze the same vid ... frame-by-frame ... to compute the actual SPEEDS (plus or minus the error range) of each vault part? Subjectively, we can guess at these speeds, but objectively, it will help to KNOW these speeds. i.e. any COMPUTED speed is probably more accurate than using the naked eye ... especially for beginners and intermediates.

These speeds will NOT be to 2 decimal places of accuracy. They might not even be accurate to a single decimal place. But if we can at least compute these speeds to the nearest m/s (meter per second), that's better than just eyeballing it ... isn't it? Remember, the target "market" for this is beginning and intermediate vaulters and coaches ... the masses ... not for elite vaulters and coaches!

The quality of the data can be improved by simply buying better cameras, and using more sophisticated software tools. The same process, just better tools ... for more accurate data. My focus is on the process.

Elite athletes and coaches can always buy better cameras and tools, and over time, prices will come down and technology (quality) will improve. It always does. But even so, there's no software tools yet (not even Dartfish) that does exactly what I'm proposing here.

BTW, by posting here, I'm already putting this into the public domain, so no one will be able to patent this process. It's free. :yes:

I do have a plan! Still unproven ... but I think it will work! ;)

I made some progress on this yesterday, and will describe this progress in my next post.

Kirk
Run. Plant. Jump. Stretch. Whip. Extend. Fly. Clear. There is no tuck! THERE IS NO DELAY!

User avatar
KirkB
PV Rock Star
Posts: 3550
Joined: Mon May 19, 2008 6:05 pm
Expertise: Former College Vaulter; Former Elite Vaulter; Former Coach; Fan
Lifetime Best: 5.34
Favorite Vaulter: Thiago da Silva
Location: Vancouver, BC, Canada

Re: Video Metrics - Vid Analysis Data Points for PV Performances

Unread postby KirkB » Mon Mar 16, 2009 12:14 am

Here’s what I did on this yesterday …

The first problem I encountered is that most professional-quality YouTube vids of elite performances are recorded in Flash Video (extension .flv) format. The default player for these is Adobe Media Player. However, AMP doesn't have the frame-by-frame arrow-key controls that the QuickTime Player. QTP uses the extension .mov. So we have to convert the video files from .flv to .mov.

There's several free converters available. I tried the one at http://www.cucusoft.com, which is a free trial download, and runs on FireFox browsers. The main drawback is a nag watermark "Trial Version" in the middle of the vid. Luckily, that doesn't actually interfere with the functionality of the vid for our purposes ... it's just a visual annoyance. It's perfectly legal to use the trial version of this software ... in fact, the publisher encourages it, in the hopes that it will lead to purchases of the Cucusoft YouTube Mate FLV_to_MOV converter product.

I’m also trying the Moyea YouTube FLV Downloader and related YouTube Converter (available here: http://www.moyea.com/ ), which doesn’t have a nag message, and runs on Windows Internet Explorer browsers. My thinking is that IE might be a little more ubiquitous for most of us, and with the 2 products coming from Moyea (and in the same work-flow), it might save time in the download/conversion process. The converted vids also don’t have a nag watermark on them. I’ll be taking Pogo’s advice, by comparing these free software tools carefully for differences in vid quality. But let’s get the FUNCTIONAL PROCESS working before we worry too much about QUALITY. That’s an optimization that can come later.

Luckily, most amateur vids are already in QuickTime format. They’re played on YouTube with their built-in player (not the Flash Video Player from Adobe), but when you download them, they play in the QuickTime Player format, without any conversion. So the FLV to MOV conversion process is only necessary (it appears) for professionally produced vids … usually of elite vaulters.

Now … I'm experiencing some problems with the quality of the vids. And I don’t mean the quality of the conversion process, I mean the quality of the vids … by amateur videographers … and even some of the so-called professional ones. The camera angles are sometimes bad ... or they don't vid the entire vault ... or people blocking the view of the camera ... or the body parts are hard to distinguish, because they're blurred, or because they blend into the background ... or twins are waving white flags … (Ha! Ha! Pogo … not really, just kidding! :D)

As already discussed, there's an issue with deciding what the best data points are. If you compare Bubka to Hooker to Colwick to Whitt, their techniques are vastly different, so it's a challenge to find common data points in all their vids. This is where DJ’s idea of “thighs together” will be more consistent than “just before the pole hits”.

I'm proceeding with just the 1997 Bubka World Championships as “the reference vid” for now. This is when he cleared 6.01 by a mile! We’ll start there, and work our way down thru the stratosphere!

If all this sounds overly complicated, remember that we're struggling thru this NOW, just to define the process. By the time we expect the masses ... vaulters and coaches ... to follow this process, it will (hopefully) be reduced down to just a few ;) simple steps. If I think back to ten, twenty, or thirty years ago, I never would have guessed that vaulters and coaches would be filming their own vids, publishing them THEMSELVES on something called "YouTube" and sharing them with fellow vaulters and coaches world-wide ... instantaneously!

It's amazing what we can do with a little technology these days! Stay tuned.

Kirk
Run. Plant. Jump. Stretch. Whip. Extend. Fly. Clear. There is no tuck! THERE IS NO DELAY!


Return to “Pole Vault - Advanced Technique”

Who is online

Users browsing this forum: No registered users and 44 guests