Merged Discussion of femr's video data analysis

But I have to keep those guards on until I examine the data in detail, which I haven't yet, or that will be turned into "So you finally admit that..." which isn't true.
You've demonstrated in multiple ways why, as far as I'm concerned, you're not particularly qualified to comment. Being critical without even bothering to examine first. Pfft. Critical for critical's sake.

And I haven't examined what NIST say in that respect in enough detail yet either.
Again, without doing so you show your opinion to be based upon incredulity. Pfft.

As far as I know you have not yet given static point region information for cam #3.
That's because you've not bothered to look.

I doubt I will be able to judge the early motion without information about static point region in cam #3.
The weight of your judgement is not particularly interesting. Plenty of information and plots of cam#3 static region data are contained within this thread.

Read Appendix C in detail before you waste any more of my time hand-holding you through detail presented years ago.
 
How about your acceleration plots ? Or W.D.Clinger's ? Or tfk's ?
As for mine, they definitely don't prove over g either. As for W.D.'s, probably same case, as he bases it on your data. As for tfk's, I don't know, I'd have to ask and I'm not really that interested.


...or NISTS ? (Remember, their linear regression is over-g, and their function for velocity, if derived to acceleration, also includes a long period over-g)
Yes, with the understanding that it is a rough approximation based on an arbitrarily chosen curve, and also that there is no error analysis for that derivative. In particular, NIST don't claim over g.


Perhaps you'd care to explain how the same general trend profile emerges from multiple SG plots by multiple persons, multiple poly fits, multiple symmetric difference plots, ... ?
I only see the same overall shape, but the vertical scale isn't all that clear.


Error and noise all biased in one direction to artificially increase acceleration with multiple methods ?
Multiple methods? How many methods of measuring the equivalence between pixels and length have been employed in these graphs? I thought they were ALL (but NIST's) based on your equivalence table of 100 px = 344 ft.


Since when has over-g been the bee in your bonnet ? :rolleyes:
Since I started recently to dig into the details of your analyses.


You think their is an error, go prove it.
I don't think there is an error, except of logic. I think you haven't provided enough proof to support the claim, and that burden of proof doesn't correspond to me.


You've demonstrated in multiple ways why, as far as I'm concerned, you're not particularly qualified to comment. Being critical without even bothering to examine first. Pfft. Critical for critical's sake.
That's dishonest. I was critical because you hadn't provided enough information for me to evaluate the claims. You are now focusing on the fact that I have not yet evaluated your newly provided data (and thus not given any conclusion based on it yet, but instead used terms like 'early motion, if present...') and using that to attack my previous assessment based on failure to produce that data. But your criticism doesn't hold water as you haven't yet provided static point location information about camera #3 traces.


The weight of your judgement is not particularly interesting. Plenty of information and plots of cam#3 static region data are contained within this thread.
In which of the 2,800+ posts? Certainly you haven't recently posted that information. A search for 'static point' was unfruitful.

I have the plots, but I don't have an indication of what region they were taken from, which is what I'm asking.
 
As for mine, they definitely don't prove over g either.
What I'm saying to you is about derivative generation methods. Numerous different folk using different methods and procedures to generate derivatives...all ending up with similar trend and magnitude for acceleration.

So, what you should be asking about is basic position/time data.

As for W.D.'s, probably same case, as he bases it on your data.
He does indeed, as do you. You are, of course, more than welcome to generate your own. Full details on how to do such properly have been provided many, many times.

Variance data has been provided for the Dan Rather displacement/time data.

"Error analysis" to the same rigour as NIST has been provided for the Dan Rather displacement/time data.

Cam#3 data is significantly higher fidelity, though of course suffers from significantly increased perspective issues.

For that reason I have not provided derivative views from cam#3 data. Your derivative views should therefore focus on Dan Rather data, not cam#3 data.

In particular, NIST don't claim over g.
So where is your lengthy discussion about why NIST did not do such ?

Where is your lengthy discussion about NIST stating 2.25s AT freefall (when in fact they stated acceleration to THREE decimal places which was itself over-g) ?

I only see the same overall shape, but the vertical scale isn't all that clear.
Why not ? Axes on a graph are as clear as each other. Can't read sideways ?

Multiple methods?
Yes. Various smoothing methods, various methods to generate derivatives.

How many methods of measuring the equivalence between pixels and length have been employed in these graphs?
HAHAHA. Nice goalpost shift. Running out of quibble room eh. lol.

One for Dan Rather data (though tfk for some bizarre and probably inept reason managed to produce data with a slightly different vertical scale).

I have provided both raw pixel data and data enabling conversion into real world units.

If you, W.D.Clinger and co have been too lazy to generate your own that's entirely up to you.

If you think you can produce more accurate scaling metrics, be my guest.

Of course the scaling metrics for cam#3 viewpoint differ.

I thought they were ALL (but NIST's) based on your equivalence table of 100 px = 344 ft.
See above.

I'll add that the base elevations are those used by NIST.

As I said, if you have an issue with the scaling metrics, get busy, generate "better".

That you lack "confidence" in them years after they were generated, cannot find any of the discussion explaining how they were generated, can't find discussion where eror and accuracy of them was discussed...not my problem.

Since I started recently to dig into the details of your analyses.
Step 1. Read the entire thread in detail. Twice.

Then find all it's offshoots. Read them. Twice.

I don't think there is an error, except of logic.
Dear me. Of course there's error in the data. Pretty noisy. +/- (a)(x)n (kinda). Thing is... +/- ...what are the odds on all error being in one direction ? What's to say the error is not UNDERESTIMATING the acceleration ? :)

I suggest the use of smoothing, averaging, .... should help you come to a reasonable answer.

I think you haven't provided enough proof to support the claim, and that burden of proof doesn't correspond to me.
What claim ?

That's dishonest. I was critical because you hadn't provided enough information for me to evaluate the claims.
Are you sure about that ? If it's in this thread, it's already been provided.

You are now focusing on the fact that I have not yet evaluated your newly provided data
What newly provided data ?

(and thus not given any conclusion based on it yet, but instead used terms like 'early motion, if present...')
What ?

and using that to attack my previous assessment based on failure to produce that data.
Failure to produce what data ?

But your criticism doesn't hold water as you haven't yet provided static point location information about camera #3 traces.
Dear me.

You have the data.

You have the video.

It's not hard to look at a pixel location on a video. HOW much hand-holding do you expect ? (Remember upscaling ;))

Perhaps a 2010 image with the static regions marked ? :rolleyes:

Cam#3 data details are not going to help with your "mission" to "investigate" Dan Rather data based derivative plots, are they.

In which of the 2,800+ posts?
Quite a few of them I expect.

Certainly you haven't recently posted that information. A search for 'static point' was unfruitful.
Not my problem.

I have the plots, but I don't have an indication of what region they were taken from, which is what I'm asking.
Yes you do. You have the data that tells you EXACTLY where. (Though bear in mind upscaling ;))

Please start using your noodle.


I don't do the hamster wheel here much these days.

If you choose to go back to square one and start "discussing" scaling metrics, go ahead. Been there, done that. Have fun.

Good "move" in your "game" though :rolleyes: Yawn.
 
Last edited:
Yes, with the understanding that it is a rough approximation based on an arbitrarily chosen curve, and also that there is no error analysis for that derivative. In particular, NIST don't claim over g.



This is a post by Frank Greening back in 2008 in the 9/11 Forum:

I have taken NIST's WTC 7 collapse velocity function given on page 603 of NCSTAR 1-9:

v(t) = 247.52(0.18562t)^2.5126 exp[-(0.18562t)^3.5126]

I have differentiated this equation to obtain the acceleration as a function of t. The result is a very long equation which I will not write out here, but I ask all you math geeks out there to please check this.

What I find is that the acceleration rises steadily from zero for the first three seconds of motion and peaks at a time of 3.2 seconds. Then, after 4.9 seconds, the acceleration becomes negative, i.e. the building's motion is decelerating after this time.

The amazing result is that the acceleration is above g for almost 1 second, from t = 2.7 to t = 3.7 seconds with a peak value of 10.5 m/s^2!

Interestingly, posters Einsteen and David Chandler have claimed to measure accelerations faster than g using different techniques of video analysis. Now it appears that even the great NIST is finding accelerations greater than g, so it's official!

But if we accept this result I have to ask:

What is going on?

He used to be one of the best posters on JREF.



Another poster differentiated the NIST equation and this is the result:

34gllzs.gif



To which Frank replied:

Einsteen:

Yes! You got it,...... over 34 ft/s^2 at t = 3.2 s, ........ amazing!

Makes you wonder why NIST came up with its strange linear fit to the velocity function derived from a central difference approximation and claim the slope is exactly 32.196 ft/s^2. (See Figure 12-77 on page 603 of NCSTAR 1-9).
 
This is a post by Frank Greening back in 2008 in the 9/11 Forum:



He used to be one of the best posters on JREF.



Another poster differentiated the NIST equation and this is the result:

http://i33.tinypic.com/34gllzs.gif


To which Frank replied:
So?
What does it mean? What is your conclusion? You don't have one?

LOL, what is wrong with a linear fit? Nothing. How is your book coming, was it CD or not? lol

You will not be making any sense on this post, you don't do math well. Greening was teasing you guys. You don't know Dr. Greening; he got you. lol, thanks for the comedy, you have made my week.
You can't summarize your book's conclusion.
 
For that reason I have not provided derivative views from cam#3 data. Your derivative views should therefore focus on Dan Rather data, not cam#3 data.
There's actually little basis for doing as you suggest, regarding the NW corner. The variation in distance to the camera is so small, that considering the distance constant is a good approximation. And since there's no error analysis in measurements in either Dan Rather or Cam #3 data, the vertical axes are equally unreliable in both and therefore only the overall shape is significant, making both useful for the same purpose, whatever that is. The only purpose that comes to mind is showing that NIST's linear approximation is not that bad within a margin of error.


So where is your lengthy discussion about why NIST did not [claim over g] ?
What?

I already said that that part of the report is not of any significance and I think it should not have been included. I haven't discussed that with NIST. Are there NIST members in this forum?


Why not ? Axes on a graph are as clear as each other. Can't read sideways ?
I don't think they are accurate.


How many methods of measuring the equivalence between pixels and length have been employed in these graphs?
HAHAHA. Nice goalpost shift.
Goalpost shift? I've been talking about my issues with your error analysis since long. Recently this in particular came to my attention:

Also, can you give an error analysis for your px-to-ft conversion? The spreadsheet with the Dan Rather raw data shows a whole number of pixels and a whole number of feet. I can't find anywhere where you take that into account in your graphs and how that error influences the velocity and acceleration plots, especially the latter, and you've so far refused to do any serious error analysis.

That's independent from the generation of the acceleration plots, with which I'm not yet done.


One (though tfk for some bizarre and probably inept reason managed to produce data with a slightly different vertical scale).
And is there any error analysis for any of the graphs derived from that single source of data (excluding tfk's)?

The obvious answer is no. They all lack error analysis, making their vertical axis marking unreliable.


If you think you can produce more accurate scaling metrics, be my guest.
It's not my burden of proof. Support your assertion.

I don't think there is an error, except of logic.
Dear me. Of course there's error in the data.
Dear me. Of course there's noise in the data. You're getting lost into endless arguing. This is the post framing the meaning of the word "error" that the above was a reply to:
The "claims" I have made about acceleration behaviour are quite conservative and based upon trend and general magnitude.

You think their is an error, go prove it.


Pretty noisy. +/- (a)(x)n (kinda). Thing is... +/- ...what are the odds on all error being in one direction ? What's to say the error is not UNDERESTIMATING the acceleration ? :)
With all of them based on your 100px=344ft, all of them are affected in exactly the same way.

And you still seem to make the same mistake again and again, of confusing "I don't think you have proved your assertion", which I am saying, with "I think your assertion is false", which I'm not. Stop it.


I think you haven't provided enough proof to support the claim, and that burden of proof doesn't correspond to me.
What claim ?
The claim of over-g in this case. Others in other cases, like early motion.


Are you sure about that ? If it's in this thread, it's already been provided.
I am sure that that was the reason I was critical. I could be critical and wrong. I am wrong sometimes too.


What newly provided data ?
In particular, this:

The most direct visual "proof" of early motion is, imo...

[qimg]http://femr2.ucoz.com/_ph/7/787997325.png[/qimg]

and this:

Here's a view of static region variance for cam#3...

[qimg]http://femr2.ucoz.com/_ph/7/904064319.png[/qimg]


Failure to produce what data ?
This:
Perhaps a 2010 image with the static regions marked ? :rolleyes:
It's the first time I see that image. I'm not sure where you have posted it in past but I do not recall having seen it. You'll excuse me if I don't follow every forum and thread where you post.

But it is grave, in this context:

In a reply to Oystein I said:

Yes, but I'm not too sure that [static point subtraction] can successfully filter out camera shake, for various reasons, especially when we're dealing with subtle features.

At the risk of introducing more confusion, here's an illustrative graph showing the velocity and acceleration trends from Cam #3.
You replied:
Of course not ALL camera motion can be extracted...sa static point traces themselves contain some noise. However, that noise is significantly lower than that for, say, the NW corner, for various reasons, and will result in a nett and significant reduction in camera motion effects....most definitely a whole lot better than not removing the camera motion.
I objected:
Was the static point far from the camera? Perspective can play a role in smoothing camera shake.
That's the core of my objection: perspective can play a role in smoothing camera shake. Yet in response you provided Dan Rather static point location:

Several regions were traced, at similar distance from the camera to WTC7...

[qimg]http://femr2.ucoz.com/_ph/7/759212128.png[/qimg]

(that one I had already seen, it's in page 22 of this thread, possibly more), and then the graph quoted above:

The most direct visual "proof" of early motion is, imo...

[qimg]http://femr2.ucoz.com/_ph/7/787997325.png[/qimg]
I complained:
Why do you post an image of Dan Rather? You posted a Cam #3 static point graphic. Seriously, you really have a problem communicating.

(when I said you posted a Cam #3 static point graphic, I was referring to post #2638).

But now you provide this image about Cam #3 static points, which don't hold to the principle that you stated that "Several regions were traced, at similar distance from the camera to WTC7..."

Therefore my objection holds: perspective can play a role in smoothing camera shake, and that holds true for cam #3 data. If I have enough interest, which is not too clear at this moment, when I have time, I will take a look by myself at the influence of perspective and the possibility of camera shake. Meanwhile you can also give it another try to presenting your case discarding these and other effects if you want.
 
Major_Tom, Greening is welcome to interpret NIST's graph as he wishes, which doesn't make his conclusions on the subject valid. The linear regression is one approach to obtain an approximation to the acceleration for discussion purposes. The curve is one approach to obtain an approximation to the velocity profile for discussion purposes, not for the purpose of being able to derive acceleration from it.
 
Major_Tom, Greening is welcome to interpret NIST's graph as he wishes, which doesn't make his conclusions on the subject valid. The linear regression is one approach to obtain an approximation to the acceleration for discussion purposes. The curve is one approach to obtain an approximation to the velocity profile for discussion purposes, not for the purpose of being able to derive acceleration from it.

Just to add: Given that the curve in question is below the linear fit at about 2.5 s, then roughly tangent to the linear fit at about 3.5 s -- and that the slope of the linear fit is very close to g -- obviously the curve rises at greater than g for about a second, no differentiation required. But to infer greater-than-g acceleration from the curve is to commit a fallacy of misplaced selective concreteness. There is no obvious reason to prefer the curve fit to the linear fit during NIST's "Stage 2." It's rather odd to bother to report the result of differentiating that curve.
 
Just to add: Given that the curve in question is below the linear fit at about 2.5 s, then roughly tangent to the linear fit at about 3.5 s -- and that the slope of the linear fit is very close to g -- obviously the curve rises at greater than g for about a second, no differentiation required. But to infer greater-than-g acceleration from the curve is to commit a fallacy of misplaced selective concreteness. There is no obvious reason to prefer the curve fit to the linear fit during NIST's "Stage 2." It's rather odd to bother to report the result of differentiating that curve.

I simply reproduced the NIST equation.

The problems with the NIST data were already pointed out. I don't infer anything from the NIST data other than it is very poor relative to what is currently available.
 
There's actually little basis for doing as you suggest, regarding the NW corner.
Absolute nonsense.

You're "contesting" "claims" about acceleration behaviour.

All of my "claims" about acceleration behaviour relate to the Dan Rather data, as I just reminded you I haven't produced any derivative data from the cam#3 viewpoint :rolleyes:

The variation in distance to the camera is so small, that considering the distance constant is a good approximation.
Those demanding full three dimensional position determination would appear to disagree with you, yet you never chose to quibble with them. One wonders why...

It is quite small, but it's certainly not zero, and certainly a lot higher than for the Dan Rather viewpoint...and that's without reminding you that perspective error occurs in THREE dimensions.

The only reason the NIST moire method functions is that the West edge of WTC7 in the cam#3 footage is not vertical.

Regardless, as I've said to you multiple times now...I haven't generated derivative data from the cam#3 viewpoint, so when you are discussing derivative data stick to Dan Rather.

Your constant flipping between the two separate datasets is beyond irritating.

And since there's no error analysis in measurements in either Dan Rather or Cam #3 data
Nonsense. I've already told you that displacement/time error analysis has been provided to the rigour of the equivalent NIST data for the Dan Rather data. That you've still not found it doesn't mean it doesn't exist. (And that includes inclusion of scaling metric uncertainty)

the vertical axes are equally unreliable in both
Equally ? LOL. Far from it son.

Cam#3 data is presented in pixels in case you hadn't noticed :rolleyes: (and for good reason)

only the overall shape is significant
I agree that trend shape is the important factor for derivative acceleration data, sure.

Have been saying so for a long time.

Multiple derivative methods produce similar shape and magnitude, therefore your personal gripe is your personal lack of knowledge about scaling metric uncertainty, which has already been provided.

making both useful for the same purpose
Nonsense. I have not provided derivative data from cam#3 data, making assertions about acceleration whilst mentioning cam#3 data utterly separate.

If you don't stop this vague wandering I'm going to have to highlight in tedious detail your meandering scope shifts. Not a pretty sight.

Sort out your focus.

I already said that that part of the report is not of any significance
So why are you wasting your life scritinising details I provided years ago ?

Why is it okay for NIST to omit error analysis step (x) but absolutely crucial for me to include it or you must throw out the data ? :)

I haven't discussed that with NIST. Are there NIST members in this forum?
ROFL.

I don't think they are accurate.
Which ones ?

Goalpost shift?
Yes. You're flipping between Dan Rather and cam#3 datasets faster than a ...flipping between early motion data and acceleration data faster than a... etc.

That's independent from the generation of the acceleration plots, with which I'm not yet done.
I could say you're done when I'm so bored as to stop pandering to your requests, but I'm far to nice for that ;)

Why are you repeating yourself ?

I've already told you uncertainty data for scaling metric has already been published, and "Haven't done such for derivative plots. Very unlikely to do so. You're welcome to do so yourself if you feel the need. I don't. Good luck quantifying error post SG derivative".

If your memory doesn't reach that far back, I'#ll provide you a link. Hint: It's the post immediately after you asked.

They all lack error analysis, making their vertical axis marking unreliable.
Then go ahead and perform an error analysis.

It's not my burden of proof.
You want to know. Get busy ;)

Stick a +/- (x) ft/s^2 on it if you like.

Support your assertion.
What assertion ?

You're getting lost into endless arguing.
Am utterly bored with going round and round in circles with you.

With all of them based on your 100px=344ft, all of them are affected in exactly the same way.
Then do your own ! :rolleyes:

And you still seem to make the same mistake again and again, of confusing "I don't think you have proved your assertion", which I am saying, with "I think your assertion is false", which I'm not. Stop it.
I'm not wasting any more of MY time pandering to YOUR personal views of how much error analysis YOU need for YOU to accept details.

tfk did a wonderful error analysis (yeah, right) and cam up with (iirc) +/- 80 ft/s^2.

Just apply that and you'll be fine.

Assume every point on every plot is actually 80 ft/s^2 higher than it should be and you will in fact prove WTC7 launched itself into space within a few minutes.

Job done.

The claim of over-g in this case. Others in other cases, like early motion.
Exactly my point. Your wandering assertions about "claim" are all over the place. I suggest your only issue at this time is that you haven't found detail about scaling metric uncertainty.

When you've found it, come back to me.

In particular, this
That's Dan Rather data....from 2010.

A recently posted view simply because you are seemingly incapable of understanding the blindingly simple reasons why subtracting static region data from feature motion data results in data which is primarily real motion.

The graph was generated in 2010. No-one else needed to see it that way...which doesn't say much for... :rolleyes:

Also from 2010.

This: It's the first time I see that image. I'm not sure where you have posted it in past but I do not recall having seen it. You'll excuse me if I don't follow every forum and thread where you post.
I won't excuse you when you come out and repeatedly say crap like "using that to attack my previous assessment based on failure to produce that data"...when the issue is your pesonal lack of knowledge.

The problem here is your failure to study the background for the topic you're in.

But now you provide this image about Cam #3 static points
No, I provided it in 2010, ffs.

which don't hold to the principle that you stated that "Several regions were traced, at similar distance from the camera to WTC7..."
A statement which applies to Dat Rather data :rolleyes:

Therefore my objection holds: perspective can play a role in smoothing camera shake, and that holds true for cam #3 data.
Dear me.

Have you LOOKED at the scale of the cam#3 static region variance ?

It's easy to see from the graph you just posted...

Less than +/- 0.02 pixels.

Have you LOOKED at the magnitude of the camera motion over the entire period of the video ?

It's easy to see from the graph you just posted...

Less than +/- 0.02 pixels (with a bit of a shake around the 90s mark)

Have you LOOKED at the magnitude of the difference between static region data and feature motion data ?


Tell me, is it the same order of magnitude ? :rolleyes:


You think it's okay to ignore perspective issues when it suits you, yet call +/- 0.02 pixel camera motion significant to displacement data which reaches order(s) of magnitude larger change in position ?

I'm going to write another post soon highlighting these two elements of your perspective when I have time. It makes the reasoning behind your "analyses" quite transparent. Cue something to do with femr2.
 
As an example of shifting your own standards to suit yourself, consider your opinion on perspective issues caused by the cam#3 viewpoint...

femr2 said:
Cam#3 data is significantly higher fidelity, though of course suffers from significantly increased perspective issues.

For that reason I have not provided derivative views from cam#3 data. Your [discussion of my] derivative views should therefore focus on Dan Rather data, not cam#3 data.
There's actually little basis for doing as you suggest, regarding the NW corner. The variation in distance to the camera is so small, that considering the distance constant is a good approximation.

(imagine if I had just suggested ignoring perspective correction for cam#3 viewpoint :eek:)

...yet near the end of the same post you flip your opinion...

pgimeno said:
That's the core of my objection: perspective can play a role in smoothing camera shake.

The core eh ?

It appears more linked to whether you're on your defense game, rather than your offense game.


1) Static region trace data (aka camera motion) for the cam#3 viewpoint has extremely low magnitude, under 0.02 pixels over a period of 3.5 minutes.

2) Early motion oscillations "claimed" indicate motion with peaks from 0.2 pixels (10 times larger) to 0.6 pixels (30 times larger)

To suggest that perspective issues (on 0.02 pixels for a 720*480 pixel interlaced frame size) are the core to your objection against early motion "claims" is "most illuminating".

To suggest that the effect of perspective for displacement data for the entire NW corner descent (which spans over 2 hundred pixels) is insignificant is "most illuminating".

Just a temporary blip in your viewpoint ?

Nope...

Therefore my objection holds: perspective can play a role in smoothing camera shake, and that holds true for cam #3 data. If I have enough interest, which is not too clear at this moment, when I have time, I will take a look by myself at the influence of perspective and the possibility of camera shake. Meanwhile you can also give it another try to presenting your case discarding these and other effects if you want.
Oh dear.

1) Perspective does indeed have some effect upon removal of camera motion, however, your objection does not hold as a means for you to continue "doubting" early motion data presented, for reasons clearly presented.

2) "Interest/time" based get-out clauses noted.

3) Perspective effects are constant, they don't suddenly change.

4) Camera shake is not a "possibility" it is very real, though thankfully incredibly slight for the cam#3 viewpoint, enabling wonderfully revealing plots of early motion to be presented with confidence.

5) I don't need to discard anything.

6) What "other effects" are you insinuating are being discarded ?

Stop shifting your viewpoint (no pun intended) depending upon whether you're arguing against me or defending yourself.
 
Exactly indeed.

This thread is the same sort of mess of confused detail we see with truther and troll dominated threads. Where the truthers or trolls objective is to deliberately keep discussion down in the details so there can be no progress. We "debunkers" should be capable of a far higher standard of objective reasoning. And steps one and two of that process (1) Be clear as to the aim; AND (2) Work from the context downwards progressively identifying, clarifying, defining, agreeing which bits of detail are necessary, relevant to and significant in the explanation or hypothesis.

So what is "the claim?"

The OP raised two, one has been dealt with one still carrying through. Plus we have one new one.

The "spot on topic" relevant claim which is still before us stated simply and in my words is:
1) femr's methods have errors - that aspect stated explicitly made many times; AND
2) the errors are "big enough to matter" - i.e. the errors invalidate use of the data - that aspect rarely made explicit but a must have factor without which the aspect of errors has no effect.

(The second claim in the OP was doubts about "sub-pixel" resolution. A matter that I suggest is no longer under any legitimate contention.)

However the first aspect "femr's methods have errors" has carried through explicitly - with some changes of wording irrelevant to this statement.

Standing alone that claim is a truism which is not in dispute despite multiple attempts to misrepresent it as "the claim".

All measurements of this type will have errors - simply put from two sources viz "noise" and "artefacts of the processing".

So, standing alone, the statement "femr's methods have errors" is of no value. It as been acknowledged many times by femr. In fact his whole set of processes are built around minimising the impacts of "noise" and "artefact induced errors". So much for false claims to the contrary. So much for the comprehension of those who don't recognise that.

The real "claim", which is essentially being evaded, is in the second part viz "are the errors big enough to matter"?

Recent discussion has seen obfuscation of that simple reality. Reinforced by wall of text postings, lots of foggily utilised details and outright personal attacks.

All without addressing "are the errors big enough to matter". Despite that issue being explicitly identified multiple times. There can be no genuine ignorance of it. So much for rational debate.

Now the question "are the errors big enough to matter" demands two bits of supporting evidence or definition viz:
A) Quantification of the possible scale of errors (Yes - naturally - risk managed so that the result is fail safe); AND
B) Identification of the purpose to which the data is to be put.

So that, at the risk of being b*** obvious, the range of error can be compared to the intended purpose of use to determine if the "error is big enough to matter".

And there will be multiple purposes supported by a range of different parties.

So that is the main claim before this thread. Current status is that argument has been focussed on the irrelevant aspect of "is there error?" Of course there is error. That is not the question and keeping the discussion focussed on that question is a strawman debating tactic plus a derail and evasion.

The real question is "are the errors big enough to matter"? I have made that situation explicit on several occasions. The substantive content of my statement has been ignored and personal attacks directed at me by those who obviously don't want to address the real issue.

Meanwhile a second "claim" has emerged. It is from pgimeno and the central thrust wanders around his claim that he does not accept femr's explanation of his data/measurements. Whoopy do! Note that is is a form of personal incredulity and does not address the main topic - "are the errors big enough to matter". (Which is accessible to reasoned analysis without any need for acceptance of femr's explanation.)

Meanwhile I have attempted to engage pgimeno in discussion without success. He ignores the main thrust of my explanations and looks for bits of detail he can disagree with. Both those claims which I state bluntly - "ignores argument" and "looks for nit pick details" - he admits both, using weaker language, in his most recent response to me. That response is a Gish Gallop of avoidance, misinterpretations and one bit of fabricating history. I will not be falling for "Gish Galloping".

Now, until we see claims which address quantification of errors and the purposes for which femrs data may or may not be valid - this thread will keep circling.

AFAICS all the points of detailed criticism raised recently are subordinate to that question viz "are the errors big enough to matter".

Current Status of Debate

The status of debate is that femr2 has put forward claims as to his techniques. And, despite all the noise and personal snark, femr's claims have not been competently rebutted.


So, with a bit of paraphrase:

"The Claim" is:
"femr's video data analysis is subject to error and the errors are big enough to matter."

The status of discussion is:
1) femr has made claims in several areas including:
a) his own area of interest - where he claims that the current status of his work is as good as he seeks to achieve at this stage;
b) comparisons with NIST methods where he identifies multiple aspects where, prima facie, his methods are better than NIST"s; AND
c) Application use by other persons where suitability of the achieved accuracy is for them to decide.

2) There has been a lot of discussion about the existence of errors without any quantification of error. That a tactic of evasion and a strawman argument.

3) No one AFAICS has either quantified the scope of error OR related that error to a plausible usage purpose. SO
4) The real question "are the errors big enough to matter" has not been addressed. It is still "on the table" for discussion. And, IMO, it is the "key area". (:o)

I have stated that I would find discussion of that real issue interesting. If we ever get to it.
 
Last edited:
I simply reproduced the NIST equation.

The problems with the NIST data were already pointed out. I don't infer anything from the NIST data other than it is very poor relative to what is currently available.

My comments were directed toward Greening's, which you also reproduced. I really didn't bother to parse why you reproduced Greening's comments, and I'm still not sure.
 
I have stated that I would find discussion of that real issue interesting. If we ever get to it.

If there wasn't such a huge volume of frantic blue sky syndrome going on, the useful content within this thread would be much more accessible.

It is beyond boring (and frustrating of course) to be confronted by those who have, firstly, not had to contend with the nonsense within this thread on a fairly regular basis since its inception, and secondly have not even bothered to spend their own time reading the thing before starting down what is, from my perspective, yet another spin around in the hamster wheel.

It is beyond boring (and frustrating of course) to be confronted by those who, firstly, assert that NIST (who performed no error analysis for their displacement/time data, let alone their derivative data) did not need to provide such as what they did "suited their purposes", and secondly demand that I perform undoubtedly unending study of error propogation to satisfy their arbitrary (and constantly shifting) standards that are not necessary for my purposes.

Error/noise details have been provided many times, even if those repeatedly moaning about the lack of such haven't bothered to actually "go find".

Bored.



Here y'are pgimeno, those "elusive" scaling metric details you said were never provided...

a) position error, in pixels

Dan Rather footage - +/- 0.2 pixels

b) scaling metric error, in pixels/ft (footage dependant)

+/- 1 pixel

For WTC 7 there is limited building measurement data available, so with the caveat of accepting the scant NIST provided values...

Vertical scaling metric 3.41 to 3.47 ft/pixel
Horizontal scaling metric 1.64 to 1.66 ft/pixel

Note that these are global metrics over the full distance, and do not affect the positional error metric.

Scaling Metrics for the Cam#3 footage are of higher accuracy, as the footage is of higher quality and resolution (which is why I stated to tfk at the beginning of our discussion that I'd prefer to use that footage)

c) velocity error, in ft/s (footage dependant)

There has been no agreement of noise reduction or smoothing process. Until there is *some* agreement, it's too early to state.

d) acceleration error, in ft/s^2 (footage dependant)'

There has been no agreement of noise reduction or smoothing process. Until there is *some* agreement, it's too early to state.
...would you Christmas Eve it. Told myself I'd not bother to grab them for you, but the endless incredulity is incredibly tedious.

AUGUST 2010.

Page 6 of this 71 page thread.

Didn't even bother to read the first 10%.

Took me about 2 minutes to locate.


What other base details do you think you're missing ?

If you want to determine propogated error for derivative data, you can do it yourself, based upon the methods you use to generate them.
 
I have stated that I would find discussion of that real issue interesting. If we ever get to it.
If there wasn't such a huge volume of frantic blue sky syndrome going on, the useful content within this thread would be much more accessible.
Yes indeed. I cannot resist the "analogy" - there is so much noise in the thread that it is hard to access the real data. :o

I hesitated over whether to respond to the multiple "miss the point and be proud of it" setting of pgimeno's last post. Just to trace each one of those errors of ignore then misrepresent would require me to go back at least four layers of post/response/counter response etc in order to rigorously rebut each of the ten or so falsehoods and respond positively to the two true and valuable points made. One of those potentially a step forward - a "baby" that I discarded with the fluid consequences of daily ablutions. I loath "reversed burden of proof" as the common truther tactic. But "reversed burden of re-re-re-researching long past statements" is nearly as bad esp coming from the debunker side.

...It is beyond boring (and frustrating of course) to be confronted by those who have, firstly, not had to contend with the nonsense within this thread on a fairly regular basis since its inception, and secondly have not even bothered to spend their own time reading the thing before starting down what is, from my perspective, yet another spin around in the hamster wheel...
It's part of the same "reversed burden of assembly of argument" - people hit you with one or several "bits" but expect you to assemble them into their argument. Apart from the presumption that you understand what their argument would be if they ever made it.... :boggled:

[/EndMinorAdrenalineBoost :rolleyes:

It is beyond boring (and frustrating of course) to be confronted by those who, firstly, assert that NIST (who performed no error analysis for their displacement/time data, let alone their derivative data) did not need to provide such as what they did "suited their purposes", and secondly demand that I perform undoubtedly unending study of error propogation to satisfy their arbitrary (and constantly shifting) standards that are not necessary for my purposes.
Avoiding comparison with NIST is essential to "missing the point" when your 16 Points Summary establishes prima facie that your methods are better.
...Error/noise details have been provided many times, even if those repeatedly moaning about the lack of such haven't bothered to actually "go find".

Bored.
Yes and Understood respectively. Non illegitimis carborundum.
 
Last edited:
In particular, this:

The most direct visual "proof" of early motion is, imo...

[qimg]http://femr2.ucoz.com/_ph/7/787997325.png[/qimg]

and this:

Here's a view of static region variance for cam#3...

[qimg]http://femr2.ucoz.com/_ph/7/904064319.png[/qimg]
That's Dan Rather data....from 2010.
WAIT, WHAT?

If you mislabel the graphs, don't complain that I switch the datasets.

Your constant flipping between the two separate datasets is beyond irritating.
:i:


tfk did a wonderful error analysis (yeah, right) and cam up with (iirc) +/- 80 ft/s^2.

Just apply that and you'll be fine.
I'll do:

Your claim of over-g is not supported by your analysis because the error is too big to claim a specific acceleration.

Thanks. It was useful.


As an example of shifting your own standards to suit yourself, consider your opinion on perspective issues caused by the cam#3 viewpoint...
Variation in distance to the camera is much greater between the static points you chose and WTC7 than between the top and bottom of the visible part of the NW corner of WTC7. I don't know why you say I'm shifting my standards. I've always said that perspective correction depends on the variation of distance to the camera.


Here y'are pgimeno, those "elusive" scaling metric details you said were never provided...
This counts as "not provided" to me:
c) velocity error, in ft/s (footage dependant)

There has been no agreement of noise reduction or smoothing process. Until there is *some* agreement, it's too early to state.

d) acceleration error, in ft/s^2 (footage dependant)'

There has been no agreement of noise reduction or smoothing process. Until there is *some* agreement, it's too early to state.
And this counts as "never provided" to me:
I've already told you uncertainty data for scaling metric has already been published, and "Haven't done such for derivative plots. Very unlikely to do so. You're welcome to do so yourself if you feel the need. I don't. Good luck quantifying error post SG derivative".
You are basing your claim of over-g on your acceleration graph, for which you never provided error analysis, which is what I've been saying. You keep getting lost in the details. Strawmanning seems to be your favorite technique and you have to be reminded quite often that you're arguing over something I haven't said. I didn't say you didn't publish error analysis for your raw data. And arguing against your strawmen is taking the scarce time I have lately. Speaking about irritating.
 
WAIT, WHAT?

If you mislabel the graphs, don't complain that I switch the datasets.
Ooh, look, you found a mistake !1!eleventy. I meant "Cam#3" obviously. No, the graph label is quite correct.

What makes you assume the graph label is wrong, rather than a simple brain-fart ? (my guess is you'd prefer the graph label to be wrong eh)

The pertinent point being that it's data from 2010, in response to your silly assertion... "You are now focusing on the fact that I have not yet evaluated your newly provided data".

And your response to the 2/3rd's of my post prior to that ? Ah, ignored. Convenient.

Your claim of over-g is not supported by your analysis because the error is too big to claim a specific acceleration.
1) How big is "the error" ?
2) Who is claiming a specific acceleration ?
3) What makes you assume that acceleration is overestimated rather than underestimated ?

Without you providing answers to those questions your assertion is unsupported.

What is your problem with over-g behaviour ?

1) NIST data contains multiple over-g segments. (And the entirety of their 32.196 ft/s2 linear regression value is itself over-g, with their assertions stated with no error analysis at all)
2) Chandler data contains multiple over-g segments.
3) At the start of the over-g period indicated by my data the NW corner had descend about 3ft.
4) At the end of the over-g period indicated by my data the NW corner had descend about 38ft.

As has been discussed many times within this thread, to bizarrely general agreement, over-g behaviour such as is indicated by my data, is expected.

You personally want to reject the data for your personal use...no worries. You can use Chandler's instead.

Thanks. It was useful.
For highlighting your intent, motives and "M.O.", sure.

Variation in distance to the camera is much greater between the static points you chose and WTC7 than between the top and bottom of the visible part of the NW corner of WTC7.
So ? You might want to ask yourself what information could be gleaned by tracing two static regions at different distances from the camera... ? :rolleyes: (You're clearly out of your depth. Suggest you stop.)

I don't know why you say I'm shifting my standards.
:dl:

See my previous post.

I've always said that perspective correction depends on the variation of distance to the camera.
I'm afraid perspective correction "depends upon" much more than that :rolleyes:

You are basing your claim of over-g on your acceleration graph, for which you never provided error analysis
You quite sure about that ? :rolleyes:

which is what I've been saying
The actual focus of "what you are saying" has (not so) subtly shifted as I've corrected your false assertions along the way. You may not even realise you're doing it, which is funny as hell from my perspective.

You keep getting lost in the details.
What is the text for that irony box exploding ? :rolleyes:

Strawmanning seems to be your favorite technique and you have to be reminded quite often that you're arguing over something I haven't said. I didn't say you didn't publish error analysis for your raw data. And arguing against your strawmen is taking the scarce time I have lately. Speaking about irritating.
ROFL.

I hope the more discerning readers of this "discussion" find that comment both "outrageous" and "funny" at the same time.

How dare you accuse me of... ! :rolleyes:


Again, if there are further base details you require to determine propogated error in your derivative plots, let me know.

If you do generate an actual value for error on the acceleration plot, you best make sure it's quite big, as I have about 7 ft/s2 headroom ;) (You might want to consider that a conservative estimation based upon wide window symmetric difference (rather than wide window savitzky-golay) would tot up to something around the +/- 3 ft/s2 ;))

(Not that you'll bother, of course, as your goal is not based upon interest in the accuracy of WTC7 acceleration data in the slightest, but clearly and simply attempt to "prove femr2 wrong version the umpteenth". My response is already written, btw.)
 
Last edited:
What makes you assume the graph label is wrong, rather than a simple brain-fart ? (my guess is you'd prefer the graph label to be wrong eh)
Frankly, I don't see why you say I'm switching datasets, especially given you've been doing what you criticize all the time, including the posting of Dan Rather static point location when the static points being discussed were from Cam #3. That's called double standards according to my dictionary.


And your response to the 2/3rd's of my post prior to that ? Ah, ignored. Convenient.
I didn't have that much time. Had to cut somewhere. I'm forced to be selective. You have a tendency to post walls of text. (So does Ozeco, and amazingly he complains about other doing the same...)


1) How big is "the error" ?
2) Who is claiming a specific acceleration ?
3) What makes you assume that acceleration is overestimated rather than underestimated ?
Not paying attention?
1) The error I took for that statement was tfk's 80 ft/s², as would be obvious if you looked at it in context. I was not serious using it, as would be obvious if you looked at it in context. It was an example of why error analysis matters, as would be... you know the rest.
2) You have claimed over-g. Several times during the last years. Like when discussing the number of times the corner went over g.
3) I assume nothing. Again, unsupported ≠ false. Tattoo that in your forehead so you see it in the mirror as a reminder.

All of these were responded in the post, if you looked at it in context instead of sentence by sentence. You cut the trees and then claim there's no forest.

I think you must be hitting reply before reading the post, and then reading and replying as you read each sentence, instead of reading, understanding, and then hitting reply. Which makes you miss the context completely. It's the best explanation for your MO.

*waves to your edit box* Hi, femr's edit box!


Without you providing answers to those questions your assertion is unsupported.
Done (even if the answers were already given in the context that you destroyed). My assertion is supported. It was from the beginning. Now what?


What is your problem with over-g behaviour ?
My only problem with it is that currently it has the status of bare assertion unsupported by the evidence you've provided so far.


1) NIST data contains multiple over-g segments. (And the entirety of their 32.196 ft/s2 linear regression value is itself over-g, with their assertions stated with no error analysis at all)
They don't claim it's over g. They clearly assume that any over g behavior is due to measuring error. Like the one at the 4s mark.


3) At the start of the over-g period indicated by my data the NW corner had descend about 3ft.
4) At the end of the over-g period indicated by my data the NW corner had descend about 38ft.
Something along these lines might succeed in supporting the claim. As stated, that alone doesn't. As it stands, the claim remains unsupported.


So ? You might want to ask yourself what information could be gleaned by tracing two static regions at different distances from the camera... ? :rolleyes:
At the time I posted that, you hadn't provided the static point location information. I still haven't evaluated the consequences because you keep diverting my scarce time into defending myself from your strawman attacks. You might have a valid point here, now that you've provided that information. I'll examine that when I have time.


As has been discussed many times within this thread, to bizarrely general agreement, over-g behaviour such as is indicated by my data, is expected.
Yes, including me, which doesn't mean it occurred. And before you jump the gun again, see above about what you should be tattooing in your forehead.


I don't know why you say I'm shifting my standards.
:dl:

See my previous post.
I did. Your complaints were based on points having a huge difference in distance to the camera. Let's go over this again: huge difference in distance = huge consequences; negligible difference in distance = negligible consequences. And if you had paid attention to the context of my post, you'd have seen that my suggestion regarding using Cam #3 acceleration plots included ignoring the vertical axis units and using the graph for what the Dan Rather based ones are good for: looking at the overall shape of the graph. Which, beyond the fact that it reveals a segment not as horizontal as it would be expected when comparing it with NIST's analysis, I don't see the point of.


I'm afraid perspective correction "depends upon" much more than that :rolleyes:
Maybe I should have said perspective distortion (which needs a corresponding correction) depends on distance to the camera. You usually have no problems with using the terms the way you want them to mean. Now having problems if someone else does the same? Dude, you're full of double standards.

Perspective correction methods use sophisticated techniques to find out the 3D location of the data, in order to determine its distance to the camera from a given viewpoint, and compensate for that by undoing the projection the camera has done in order to obtain a purely linear relationship, and undoing it depends exclusively on distance to the camera. In the end it all boils down to that. The 3D to 2D transformation done by a camera (or our eye) is given by:

x' = k1·x/z
y' = k2·y/z

where x' and y' are the projected point's coordinates (the pixel coordinates of the image), k1 and k2 are constants that depend on the FOV and aspect ratio, x and y are the horizontal and vertical coordinates of the point to project, transformed to the camera's reference frame, and z is the depth, or in other words, distance from the camera. There's a slight difference between both concepts (distance and depth) in that the distance is sphere-shaped while the depth is plane-shaped, but at the usual distances, the radius of that sphere is so big and the region to consider so small in comparison, that the difference is negligible, therefore using the term "distance from the camera" as an approximation to refer to depth is valid.

If it were not by z, the relationship would be purely linear, as happens with orthographic projection, given by:

x' = k1·x
y' = k2·y

And at great distances (such as that in the Dan Rather viewpoint), a small relative difference in distance implies so little perspective distortion that it's negligible. For example, the result of dividing e.g. by 5002 meters is not significant with respect to dividing by 5005 meters. Photographs taken from a very far distance with big zoom look almost orthographically projected for that reason.

The buildings you chose as static points may well be about half the way towards WTC7; that makes a 2x difference in perspective correction. Big deal, compared to the difference between distance to the top and distance to the bottom of the NW corner, which for the purpose of watching the shape of the acceleration plot, is negligible. You implicitly acknowledged that importance when you pointed to the Dan Rather static point location, saying that the building chosen was at a similar distance to the camera as WTC7 was.

Therefore, I have not switched my standards whatsoever, quod erat demostrandum.


You are basing your claim of over-g on your acceleration graph, for which you never provided error analysis
You quite sure about that ? :rolleyes:
What? You posted some variance information, which is incomplete without the propagation of error in the measurements. And you said this yourself:
and how that error influences the velocity and acceleration plots
Haven't done such for derivative plots. Very unlikely to do so. You're welcome to do so yourself if you feel the need. I don't. Good luck quantifying error post SG derivative ;)
So yes, I'm sure about that, in context. NOW you have provided a bit of error information taken from the symmetric difference method, the consequences of which I still have to analyze together with all the pending stuff. For some reason I suspect it doesn't apply to any acceleration graph you've provided that you've used to claim over-g.


The actual focus of "what you are saying" has (not so) subtly shifted as I've corrected your false assertions along the way. You may not even realise you're doing it, which is funny as hell from my perspective.


What is the text for that irony box exploding ? :rolleyes:
Indeed. I'll have to go with this one.
:i:

What you have done so far has been to take my quotations out of context and building strawmen out of the decontextualized version of them. I could only identify very few valid objections you've made, which I can't address because the amount of bullcrap, diversions, misinterpretations and strawmen in the rest takes all of my time.


I hope the more discerning readers of this "discussion" find that comment both "outrageous" and "funny" at the same time.

How dare you accuse me of... ! :rolleyes:
I'm afraid the more discerning readers will see your sentence splitting and decontextualization and see through the way you manipulate it in order to make it seem you're right.
 
Last edited:
Forgot to address this:

The pertinent point being that it's data from 2010, in response to your silly assertion... "You are now focusing on the fact that I have not yet evaluated your newly provided data".
No, that's not the pertinent point, it's just wishful thinking: you would like it to be the pertinent point in order to make you seem correct. It would be the pertinent point if you had provided it in a thread I was reading in 2010. From my point of view it's newly provided data. Again, don't expect me or anyone to pursue you to each forum and read each message you've posted.
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom