*shrug* You keep missing the word "segment" which was there since the beginning and validates my reference. Any graph with a straight horizontal line at about 32.2 ft/s² would have served the exact same purpose; even tfk's graph. Your strange way of understanding things is not my problem. Also, unlike you, I do concede when I think you're right, which in this case I don't. Your misrepresentation of the facts in order to attack me is what concerns me here. And your ad hom "you can't be trusted" is a low blow. There are many more issues of trust with you and your blunders, misinterpretations and nonstandard use of terms.You can't be trusted. Swamp the trivial point with waffle.
My definition has been steady all this time. Your interpretation is what has been wandering (read: wrong most of the times) and I have had to correct numerous times and to use big red letters and to suggest you to tattoo it in your forehead. And for the looks of it, I'm not yet done.I'm "soooo" not interested in your wandering definition of "wrong".
Again, I don't know if there's a period of over-g because you haven't proved there is. I don't care enough as to put the effort of proving it either way. I do care that you make a categorical assertion without proving it.If you don't think there is a period of over-g, prove it(And good luck with that)
Maybe. I got that idea from here:I haven't dismissed "depth" in the slightest.
I'm afraid perspective correction "depends upon" much more than thatI've always said that perspective correction depends on the variation of distance to the camera.![]()
Are you quite sure about that ?Perhaps you'd care to highlight why a point TL of frame and a point BR of frame are not affected by lens field of view effects. Perhaps you'd care to highlight why narrow field of view lenses and wide field of view lenses are identical in terms of perspective correction. Etc.
Who are you trying to "educate" ?
It's clear you are googling as you go.
Parameters.What "SG data" have I refused to provide ?
If you claimed inch precision, maybe I'd question that. I plan to give error bands for a range of credible camera positions.ROFL. If I provide you with such data you'll simply spend another year whingeing about how it was calculated. Do it yourself.
Static point location information for cam #3 was provided only several messages after I asked. There have been quite some other similar instances in which you haven't provided the data. We'll see if you provide the pixel aspect ratio information I request below.What "other data" have I refused to provide you with ?
I have ideas for all of these. In the end it all translates to an error band. Any other variable you want controlled?Splendid. Now, how are you going to go about synchronising multiple videos, managing differing scaling metric, feature resolution, differing noise levels, differing camera position accuracy, ... ?![]()
Then?That would be a wonderful "clarification", except the pixel aspect ratio isn't 1:1.
I see nothing about pixel aspect ratio. Anyway I planned to add some checks.See my previous post![]()
Doesn't mean either is correct. That doesn't prove over-g. No reason to comment on a non-proof.Also, a quick points from my earlier post you decided to ignore...
The useful output from all the various trend views is simply trend shape and magnitude. All methods presented show similar trend shape and similar magnitude. Peak acceleration between both poly and both SG plots are within a few ft/s^2.
(How long does it take you to write a post ?! Approaching 2 hours...)
I'm not the only one doing it, not even the one who initiated the silly criticism on the choice of example.LOL. No misrepresentation in the slightest. Perhaps a difference in interpretation, however, it's not me going to laborious and pedantic levels to try and make some kind of point out of a most trivial "difference of opinion". When people do that it speaks volumes.
Do you really want me to list all the times where you've interpreted that I said there was no period of over-g (or acted as if you interpreted that), when I've said no such thing?ROFL. Get a grip man. By all means list these "corrections".
How many times do you fall in the same mistake?
Unsupported ≠ false
That's not the most simplistic, at least not in the sense that it treats the raw data directly without further processing that can introduce error. I've done a quadratic fit over the raw data in that interval instead, and got this out of QtiPlot:Have you managed to perform the most simplistic determination of acceleration between the 12.5s and 14.5s mark yet ?
Polynomial Fit of dataset: RawPositionData_2, using function: a0+a1*x+a2*x^2
Weighting Method: No weighting
From x = 1.251251251251000e+01 to x = 1.449783116450000e+01
a0 = -2.638454150958290e+03 +/- 1.791304696042726e+01
a1 = 4.283697269612591e+02 +/- 2.656166316734676e+00
a2 = -1.738928555529672e+01 +/- 9.832085836278116e-02
--------------------------------------------------------------------------------------
Chi^2/doF = 1.034921189393713e-01
R^2 = 0.999831069675003
Adjusted R^2 = 0.999826700787287
RMSE (Root Mean Squared Error) = 0.321701910064848
RSS (Residual Sum of Squares) = 12.1085779159065
---------------------------------------------------------------------------------------
You've proved such to your satisfaction AND stated it publicly as a fact, not as your opinion.I've proved such to my satisfaction.
Let's define concepts before we get lost in semantics.My statement...
"I'm afraid perspective correction "depends upon" much more than [variation of distance to the camera]"
...is quite true, standing, sitting or doing cart-wheels.
By the definitions above, lens distortion doesn't affect perspective correction. Maybe you're using a different definition of perspective correction. State yours.Lens distortion affects perspective correction, which is what you are proposing to do.
Not if you did the corresponding error analysis.If I didn't claim inch precision, you'd complain ad infinatum about placement accuracy.
You miss context very often for that reason.Er, riiiight. Hmmm. Haven't read ahead...
Thanks.10:11.
Aspect ratio.About what ?Anyway I planned to add some checks.
Code:Polynomial Fit of dataset: RawPositionData_2, using function: a0+a1*x+a2*x^2 Weighting Method: No weighting From x = 1.251251251251000e+01 to x = 1.449783116450000e+01 a0 = -2.638454150958290e+03 +/- 1.791304696042726e+01 a1 = 4.283697269612591e+02 +/- 2.656166316734676e+00 a2 = -1.738928555529672e+01 +/- 9.832085836278116e-02 ...
The acceleration is twice a2, 34.8. Closer to a proof but still not there.
Yawn. Only "corrections" in your own head I'm afraidI don't feel like listing all of my repeated corrections...
Simply replication of NIST's methodsThat's not the most simplistic, at least not in the sense that it treats the raw data directly without further processing that can introduce error.
Interesting to compare the 35.0 from symmetric difference (13) to the 34.8 from a second order polynomial.The acceleration is ... 34.8
Then carry on with your analysis until you are satisfied.Closer to a proof but still not there.
Correct.You've proved such to your satisfaction
Where ?AND stated it publicly as a fact, not as your opinion.
No, let's start by not forgetting that the point of my response is to correct your clearly false assertion.Let's define concepts before we get lost in semantics.
Yawn. Have you just discovered "google" ?Perspective correction...
You're looking to perform perspective correction on real-world footage of WTC7, for "whatever" reason.The process of undoing perspective distortion.
Which, bearing in mind video cameras have lenses...Lens distortion: Distortion caused by optical side effects derived from the geometry of the lenses, such as barrel/pincushion/complex distortion and chromatic aberration.
Undoubtedly, however they are far from the point, and leading away from your repeated buffoonery...Do you have other definitions?
Who are you talking to ? [points, laughs]State yours.
pgimeno said:Which as it stands, is a false statement.femr2 said:I'm afraid perspective correction "depends upon" much more than thatpgimeno said:I've always said that perspective correction depends on the variation of distance to the camera.
Irrelevant.In a camera obscura
Correct.the lenses also introduce lens distortion
How do you separate the two when you have no information about the lens ?Lens correction must be applied before perspective correction
ROFL.There is one single variable to consider for perspective correction: depth.
LOL. What happened to your one variable ?Finding depth is a laborious work in which many other variables come into play; once you have it, it's all you need.
Irrelevant.In an orthographic projection
LOL. Good luck determining any of those values with any accuracy. Jeebus.In a perspective projection, distance between two points can be determined solely by depth of each point, projected position, angle with the camera plane and pixel metric information. The latter can be derived from reference points at a known depth, angle, and distance between them.
And all sorts of other sources an'all.Technically there's another factor to consider besides lens and perspective distortion, namely atmospheric distortion
State yours ? Again, who do you think you're talking to ? LOL.By the definitions above, lens distortion doesn't affect perspective correction. Maybe you're using a different definition of perspective correction. State yours.
You are suggesting missing context by not reading ahead ?pgimeno said:You miss context very often for that reason.femr2 said:Haven't read ahead...
You're welcome, however, making incorrect assertions, requoting them with corrections, without even BOTHERING to look up the PAR of 4:3 standard NTSC is pretty pathetic. I shouldn't have to give you these details, and you shouldn't be making such primitive mistakes. The problem is that you are clearly very inexperienced in the video arena, both in terms of video data and treatment.Thanks.
femr2 has said that error is bigger in the pre-release stage. Some correlation with time is possible. Video compression algorithms can do that. Probably not significant enough in this case.The inference depends on the data. However, given the tightness of the fit, it seems that random (iid) measurement error couldn't account for the apparent >g acceleration. The error would have to be substantially correlated with time -- which certainly is logically possible. Do you agree, more or less?
How do you correct perspective distortion with lens correction formulas?How do you separate the two when you have no information about the lens ?![]()
Who said accurate? You're Mr. 571.1364±8.2345, not me.You're going nowhere with an attempt to perform accurate perspective correction without dealing with lens distortion.
The problem here is obviously semantics. I asked you to define your terms for that reason, even if you've refused.Lens distortion affects perspective correction, which is what you are proposing to do.
And what about you stop assuming I don't know what I'm talking about, when I do? The only thing I googled is pixel aspect ratios, to know whether that was common. I haven't yet checked the resolution of the video.You'll get there...if you stop assuming you know what I'm talking about, when you don't...and stop assuming I don't know what I'm talking about, when I do. Seemples.
Yes. Sometimes you even split sentences in the middle and your reply doesn't make sense given the context of the whole sentence.You are suggesting missing context by not reading ahead ?
Straw man. I have not made any incorrect assertion regarding aspect ratio. Inventing mistakes is pathetic.You're welcome, however, making incorrect assertions, requoting them with corrections, without even BOTHERING to look up the PAR of 4:3 standard NTSC is pretty pathetic.
How do you apply a lens correction formula without information about the lens ?How do you correct perspective distortion with lens correction formulas?![]()
LOL. Your purpose is to improve the cam#3 data. That data is already sub-pixel accurate in nature (ring any bells ?)Who said accurate? You're Mr. 571.1364±8.2345, not me.
I disagree.The problem here is obviously semantics.
Incorrect.
As I said...
The kind of motion already extracted from videos is well sub-pixel in nature, and contains some noise.
The result of determining full three dimensional position (in other words correcting for perspective) for the trace data in question would be very minor, and an almost linear deskew. As the fidelity of the data is already sub-pixel, and the Dan Rather viewpoint is almost orthogonal, the resultant change to derived acceleration trend plots would also be minor. Perhaps a slight change in magnitude, and that's about it.
If you don't understand, take offers to explain to you in more detail.
Calling my statement "Meaningless technobabble" does you no favours.
Tagging onto tfk's nonsense is also a bad idea.
Hey, if you want to look at perspective correction for the cam#3 viewpoint, great. I didn't go there as it is a whole can of worms, and imho is likely to add noise to existing data rather than reduce it. Good luck.
You can't determine depth during descent with one camera view. You'd be better off determining camera-object (wtc7) orientation during static motion period and performing a simple geometric correction.
Good luck with that. Remember to include all the other inherent noise sources on top.
Say wha'"Lens distortion affects perspective correction".
You parse that as: "Lens distortion affects the image that you need to apply perspective correction to".
I parse that as "Lens distortion is a variable that intervenes in the process of perspective correction".
The former is true. The latter is false. Both of us were right.
In what way can you specifically identify your own error and then declare you're right ?I don't assume it. I respond to it when you prove it to be the case.And what about you stop assuming I don't know what I'm talking about
a) Picture Aspect Ratio is not 1:1.Straw man. I have not made any incorrect assertion regarding aspect ratio.
No ! There is an increase in noise around the release point. I already told you to have a look at the actual Dan Rather data to see it for yourself.femr2 has said that error is bigger in the pre-release stage.
Do you mean will take a long time ?I still have to examine the conversion from pixels to feet in more detail, but that can take long.
Wrong. My purpose is to make it usable. There's little point in having it there if it can't be used.LOL. Your purpose is to improve the cam#3 data. That data is already sub-pixel accurate in nature (ring any bells ?)
Same with exceedingly high accuracy.Without exceedingly high accuracy you're simply going to add noise to your target data.
My father was a photographer. I learnt a lot from him. Again, no need to google concepts I'm very familiar with. Also, I'm quite familiar with 3D graphics.Your problem is that I am speaking from the perspective (pardon the pun) of "what you are intending to do to real-world WTC7 footage, and the obstacles which such presents", and you are getting yourself all tied up in googling "camera obscura" in a vain attempt to "prove femr2 wrong version the umpteenth".
It's not my own error. My interpretation of your words was valid. Within that interpretation I was right. Within your interpretation you were right. I've done the effort to identify the issue and concede. You keep throwing **** at me without conceding anything.Say wha'In what way can you specifically identify your own error and then declare you're right ?
Which I didn't assert it was, thus putting in black and white that you were strawmanning... again. And again, you don't concede; you never do.Straw man. I have not made any incorrect assertion regarding aspect ratio.
a) Picture Aspect Ratio is not 1:1.
b) Pixel Aspect Ratio is not 1:1.
ETA: Also, it assumes 1:1 aspect ratio, which may or may not be correct.
- A reference point at a known height, in order to match both videos. Possibly two, depending on whether the aspect ratio is not 1:1
(Clarification: 1:1 alludes to the pixel aspect ratio, not to the image aspect ratio.)
Which is the position I have stated several times to the consternation of those who don't want to address the topic. Viz "purpose" and "scale of the obstacles" AKA errors....Your problem is that I am speaking from the perspective (pardon the pun) of "what you are intending to do to real-world WTC7 footage, and the obstacles which such presents",...
apparently - here IMO is a reason why....One gets the impression you don't want to give Mark a straight answer...
...and yes, even thoug I've clearly marked them as sample, draft fits, and not perfect yet...at that time...the sample femr2 fit is still much better than NIST's.
Think ? I know it has little effect. Comes with having the models available and the facilities to move things around at will, including lens parameters. A distinct advantage.
As I said, I suggest you continue your, no doubt never-ending, "critique" in a more appropriate thread.
@femr2: Isn't it time to cut your losses with this rolling "Gish Gallop"? [...]
The two ongoing discussions in this thread are:
1) the main topic question which can be simply stated as "Do femr's video analyses produce data which is accurate enough to be useful?"
The claims that femr's data is not accurate enough have degenerated into a pair of strawman arguments buried in a fog of personal attacks and a couple of Gish Gallops of untruths.
However if that discussion is to progress legitimately it demands quantification of the errors inherent in the analysis methods. And Mark's question starts to head in the direction of "quantification" which doesn't suit the tactics of "prove femr2 wrong version the umpteenth". (Neat phrase that one.)
Meanwhile the second discussion is:
2) pgimeno's tactic of limiting his acceptance of femr's analytical data to the requirement that femr must explain it to pgimeno's satisfaction. Plus a rolling "Gish Gallop" to make sure we all forget what the real issues are....
I don't think it is intentional but I prefer to stay away from considerations of personal motivation for reasons which apply generally to all threads and are specifically obvious in this thread. Viz where does personal antagonism take over from reasoned rational discussion.I don't think that pgimeno is attempting to execute a Gish Gallop...
Undoubtedly true but my preference is to clarify the way through the fog rather accept that is often happens and condone the fogginess....I think a lot of the discussion has been chippy and off-point in a way that even professional discussions too often become, and that internet discussions may be especially prone to. Every imprecision or ambiguity in language can become an ongoing point of contention....
Slightly disagree there...."isn't inherently an intended tactic" Lack of intent does not preclude tactical effect as you rightly observe:Asking detailed questions about other researchers' methods isn't inherently a "tactic,"
I've been monitoring this thread to see if it returns to reasoned discussion of what I see as the primary issues viz "Do femr's video analyses produce data which is accurate enough to be useful?" I am satisfied that femr has made a prima facie case and that no one has so far addressed or rebutted the key issue of "...accurate enough to be useful". My identification of that concern has led to several outbursts of personal attack directed at me - not just the ones directed at femr. Whilst it is easy to make the "it isn't intended - it often happens" assessment that you make - and I comprehend the perspective - such is not the innocent history of this thread....but it can be -- and even with the best of intentions all around,...
There were two objectives being discussed over recent weeks as I identified in my post. We now have a third one which is a distraction from the real issues of differing perspective.the discussion can easily veer off point. This one is pretty far gone. So, yes, it would be great to focus on quantifiable issues...
The first step should be to decide what the aim of the discussion is. It seems to be wandering all over the place with the only constant theme being "prove femr wrong". pgimeno some time back clarified his own objective as (words to the effect of) "Get femr to explain his methods to my satisfaction". That aim specifically precludes pgimeno addressing the actual measurements - which is why I have commented on it a couple of times - and withdrawn from discussion. That aim is of no interest to me.My impression -- it can only be that, because I really don't have time to replicate analyses, nor do I have specific competence in this area -- is that in the specific question of average acceleration of the reference point over approximately a 2-second period, the results do not fundamentally depend on analytical method (how S-G is applied) or random "noise" in measurement. It could be sensitive to systematic measurement error.
Whilst I comprehend your suggestion it involves:...Perhaps the best course is for pgimeno to see if he can replicate the basic result without using femr2's data....
You are not alone - it is the fundamental problem with this thread and many more. "What the **** are we talking about?" - illustrated I think by "When up to your arse in alligators it is easy to forget that the objective was 'drain the swamp'." Also references to "forests and trees" come to mind....I'm not sure how much the answer to this question matters,...
Yes. Be clear where we are trying to head and clear as to the route we are attempting to traverse.... but as long as we're supposedly discussing empirical questions, it sure would be nice to do that.
I have done it. Method and results will be published soon.How do you apply a lens correction formula without information about the lens ?![]()
Indeed not, thanks. And I think your diagnosis of the issues is pretty accurate.I don't think that pgimeno is attempting to execute a Gish Gallop.
I might end up doing that, but currently my focus is on entirely different matters, namely perspective issues for both the plane and WTC7. My remaining point of contention is pixels-to-feet conversion, something I still need to look at.Perhaps the best course is for pgimeno to see if he can replicate the basic result without using femr2's data.
I might end up doing that, but currently my focus is on entirely different matters, namely perspective issues for both the plane and WTC7. My remaining point of contention is pixels-to-feet conversion, something I still need to look at.
While I appreciate the thoughts above, they are in and of themselves evidence for why these topics should be in the Science and Math sub-forum.
Good idea. I think my previous fit shows almost conclusively that the issue is not the noise; the only thing I still need to do to be sure is a least squares fit that fixes the second degree term of the polynomial to -16.1, and plot it to see how plausible it is compared to the -17.4 one. I suspect it won't cut it and that will definitely prove that the problem is not the noise. I already implied earlier that something along the lines of a fit that doesn't depend on error propagation might be successful, namely:I'm definitely not invested in any particular vision of the sequence in which you should work. I will comment that it isn't obvious to me whether and how this particular focus relates to the discussion of acceleration, or just where you think that discussion stands.
Here's another half-baked thought, for what it's worth: A cynic might construe that you are changing the subject because you think you can more easily criticize femr2 on some other subject. If you were to summarize some points of agreement between you and femr2 on the acceleration topic -- as well as unsettled questions on that topic -- that might contribute to recentering the discussion on empirical inquiry.
Something along these lines might succeed in supporting the claim. As stated, that alone doesn't. As it stands, the claim remains unsupported.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.
1) I'm now more convinced about early motion. The trace data was not enough for that; movement seen in the video was clearer.Exactly what is it that you are trying to prove ?
Exactly what purpose do you have for the data ?
What exactly are you trying to achieve ?
2) You have claimed that the acceleration went over g. I don't think you have proved that.
3) I am defending a claim I made and you have attacked, namely that the issue with perspective is distance from the camera (depth).
4) I think I am in a position to undo perspective effects from Cam #3. That would enable usage of that data which currently can't be taken enough advantage of.