The 9/11 Forum

Trippy wrote:

David B. Benson wrote:Trippy --- Pixel "height" is an angle.

Huh?

-Confused.

Guess DBB means point-source lense effects. Curvature of frame from focal point.

Negated somewhat by *infinite* focus on lenses, and modern lens production techniques, but if I understand correctly, a fair point....though so miniscule that adding it in on the data will be pointless. Far bigger sources of noise/error are present.

A pixel at the camera subtends some horizontal angle and vertical angle through the lens system. Measuring pixel differences, say vertically, is a measure of the angle between those pixels.

femr2 --- No, I'm not writing about that. When measuring pixels, say down from the top of the Sauret video, so at 1.6 km away, one is measuring an angle. Determining the number of meters is then a matter of knowing the horizontal distance and the number of radians subtended by each pixel. Then taking the tangent of the angle gives the vertical distance.

Elementary trigonometry, but required here.

David B. Benson wrote:A pixel at the camera subtends some horizontal angle and vertical angle through the lens system. Measuring pixel differences, say vertically, is a measure of the angle between those pixels.

femr2 --- No, I'm not writing about that. When measuring pixels, say down from the top of the Sauret video, so at 1.6 km away, one is measuring an angle. Determining the number of meters is then a matter of knowing the horizontal distance and the number of radians subtended by each pixel. Then taking the tangent of the angle gives the vertical distance.

Elementary trigonometry, but required here.

This is precisely what I was talking about

An What we have is a measure of how much the camera moved, and we need to translate that into how much that camera movement makes the tower move.
[(Foreground vertical displacement)/(Foreground Distance)] = [(WTC1 Vertical Displacement)/(WTC 1 Distance)] is what I was thinking of.

Although, if i've understood what you're saying correctly, a 1 pixel displacement of the foreground object due to camera vibration will result in a 1 pixel displacement of WTC 1 due to camera vibration?

David B. Benson wrote:femr2 --- No, I'm not writing about that. When measuring pixels, say down from the top of the Sauret video, so at 1.6 km away, one is measuring an angle. Determining the number of meters is then a matter of knowing the horizontal distance and the number of radians subtended by each pixel. Then taking the tangent of the angle gives the vertical distance.

Elementary trigonometry, but required here.

We're talking about a similar thing, just in a slightly different way, or possibly at a different scale.

Angular distortion extends from the focal point, normally (but not always) the centre pixel of the frame. Depends upon the lens.

What I mean by manufacturing of lenses being of relevance is that depth-of-field effects are taken into account in CCD's to minimise the fish-eye effects (which are an extreme case of such distortion.)

I suppose that could introduce, say, a 1m distortion, but almost impossible to determine whether it has been pre-corrected by the lens/CCD. Looking for sub-pixel curvature of long linear features would be the best way to quantify it I rekn...

---

For the camera wobble, distinguishing between planar camera movement and angular movement could be very difficult indeed, though I agree that it could be looked into...

Still think it should be relative to the centre-of-frame.

...

I've tried amplifying the foreground movements on application to the antenna trace, and it does improve things. Trial and error at the mo, but I'm using a 1.22 multiplier at the mo and it looks prety good. Will look at bolting some trig in there

femr2 wrote:I've tried amplifying the foreground movements on application to the antenna trace, and it does improve things. Trial and error at the mo, but I'm using a 1.22 multiplier at the mo and it looks prety good. Will look at bolting some trig in there

The easiest way would be if you can find the distance to the forground object, and go from there.

Trippy wrote:

femr2 wrote:I've tried amplifying the foreground movements on application to the antenna trace, and it does improve things. Trial and error at the mo, but I'm using a 1.22 multiplier at the mo and it looks prety good. Will look at bolting some trig in there

The easiest way would be if you can find the distance to the forground object, and go from there.

Hmmm. Might have been hasty on stating the multiplier improved things. 1.0 does seem to be pretty good in itself.

The gains of going to trig are probably exceedingly minor, but I'll still take a look. (Again, it would assume rotational movement of the camera *during* the descent of the antenna. Camera movement is very minor during that time.

First step will be to look at the most severe wobbles and see how much difference there is between the near building and the far tower. That will allow me to determine the properties of the movement a lot more clearly.

I've tried multipliers up to 3 and the effect on the running average graph is almost zero still..

Well now...

Tried a comparison between the near and far wobble data...



Very rough *fit* but pretty similar.

SynthEyes is excellent...

Will think about what do to with the difference data.

Observation would suggest that the initial vertical wobble was not rotational, and indicative of physical up-down movement, followed by subsequent slight rotational wobble afterwards.

Interesting.

Building shake. Hmmm.

femr2 wrote:Observation would suggest that the initial vertical wobble was not rotational, and indicative of physical up-down movement, followed by subsequent slight rotational wobble afterwards.

Interesting.

Building shake. Hmmm.

How many stories was the sauret footage shot from?

I keep thinking of the buildings as being a bit like a bunch of tuning forks on a sheet of plywood, or maybe clocks on a wall - especially if they were also built on the Schist underlying the area.

Then again, loosely pack soild amplify vibrations, so...

I recently did another rip, with femr2's assistance (though I've not availed myself of all of the advice just yet), and the marker I used to synchronize with the previous rip was frame 655. This is the last frame in which the falling object (it could be a person) associated with the big camera shake is last seen plainly at the bottom of the frame. In the next frame a slight remnant is visible.

If the other rips go back that far (surely femr2's does), then that's the best thing I've seen in a casual inspection. Won't work for other videos, of course, and femr2's time resolution has to be factored into the synchronization...

An aside: the minutiae of some of this work slays me. That I had to sanity check sync between two different rips from the same video, which were both default jobs using different software - and I swear I had the same starting frame both times - is nuts. And it failed check! The first internet copy was 10 frames offset from my prior rip from DVD, and the first go-round this time was offset 10 like the internet copy. I did it again to match the prior, which is quoted, so as to introduce no more confusion.

Now I'm going to check a later point to see if the last two rips agree in two places.

Frame 942 is when a nubbin (?) of flame first appears on the west side around floor 92. This eventually becomes the significant flame expulsion in that area over the next ~50 frames.

This sanity check between first and second rip passed, thank goodness.

femr2 wrote:Well now...

Tried a comparison between the near and far wobble data...

...

Very rough *fit* but pretty similar.

Beautiful. Velocity, right?

On second thought, velocity doesn't make any sense on one of the scales. There isn't really that much position change, is there?

Wait a sec, that includes some shake prior, your descent starts around frame 75+. Hmmm, thinking...

I sure didn't see anything like 10 pixels peak to peak shake in the 100+ frames prior to initiation. I mean, maybe a pixel or two max, I'll have to look.* Can you elaborate on that lovely animated graph? Your pixels are perhaps stretched x2, but not x5, or are they?


*Edit: I looked. A half-pixel peak to peak, tops, generally less. This is the time period prior to and including initiation (dish position in pixels):



Surely the camera wasn't doing 5 pixel excursions this way then that. 20+ times stretch???

OneWhiteEye wrote:Can you elaborate on that lovely animated graph? Your pixels are perhaps stretched x2, but not x5, or are they?

I can indeed. The video used is upscaled x4 with VideoEnhancer.



The fading graph comparing near and far is not velocity, but is vertical position, raw, of the camera wobble. The values in the 900 range are the tower feature pixel, but are not comparable to your pixel locations, as I use a cropped copy of the video for tracking (As synthEyes will not read files over 2.2Gb ). The values in the zero range are the near feature, which for my purposes are set such that the firsy point read occurs at 0,0. Probably just made any confusion worse, but I can lay it all out clearly later on.



For the other graphs posted, yes velocity. As I'm using a feature higher in the frame, I can track it for longer. That may be why we have not previously seen the large changes. I cross-checked the trace with the one from the bottom field of the interlace pair, and it's there too, so I'm pretty darn sure it's not a tracking glitch.



Note to self: Sort out the damn scaling so that units are in common terms (Seconds, and meters). And label yer bleedin graphs.

*Edit: I looked. A half-pixel peak to peak, tops, generally less. This is the time period prior to and including initiation (dish position in pixels):



Surely the camera wasn't doing 5 pixel excursions this way then that. 20+ times stretch???

Video upscaled x4.

I strongly encourage using pixels for the ordinate and fame numbers for the abcissa. In particular, the conversion from pixels to meters, vertically, is less easy than one might first think.