Here is another possible way to attack this error margin. Perhaps one of the video experts here can suggest some series of conversions and compressions that they think would introduce a margin of error of 4-5 pixels in an otherwise stable motion.
It doesn't take a video expert to answer that; it just takes some common sense, so I'll be glad to help you out.
First, there isn't any "error of 4-5 pixels" to be explained. When you increased the size of your images by 200%, you also increased your errors by 200%. Duh.
Second, the "error" in the original images is not really 2 to 2.5 pixels, either, even if your measurements were precise. The 2 to 2.5
original pixels would be the total difference you got between your largest and smallest per-frame travel distance, with the mean falling very close to half way between, so your measurement error from the plane's probable real position is
plus or minus a little more than 1 pixel to either side of the mean.
When I use the original images (enlarged enough to see individual pixels clearly, but without any resizing interpolation), and I try to measure the number of pixels between the edge of the building and the nose of the plane (which is a much easier and more accurate way to do the same thing you tried to do with your stabilization), for the seven frames before the plane's nose disappears behind the logo, I get per-frame travel of 13, 12, 12, 13, 11, and 11 -- one pixel to either side of the mean of 12.
But anyone who tries to duplicate those measurements will easily see where the plus-or-minus-1-pixel errors are coming from: It's completely impossible to determine the
exact pixel that marks the edge of the building
or the nose of the plane; you have to guess at both. Starting with imaging errors in the camera (noise, differences in CCD pixel sensitivity, and edge artifacts), compression artifacts were introduced in the original digitization of that image, and then more were added when it was converted to QuickTime.
Given that the error is just about what should be expected, it's absurd to suggest that those images prove that the plane is not moving at a constant speed. The incontrovertible proof that the images won't support any greater accuracy is to simply look at the edge of the building and note that the imaging and compression errors make what we know is a straight edge appear very ragged within a band about 3 pixels wide
and changing from one frame to the next!
The only thing you have produced here is a very nice graph of the measurement errors. Since those errors are just about what should be expected, your study actually lends support to the theory that it's a real image of a real plane moving at a constant velocity. ("No-planers" will hate you for that if they ever understand it, but that's probably not a problem.)
(Edit to add: The thing that really cracks me up about the video analysis done by "no-planers" is that they persistently take video flaws and artifacts that should be there in real videos, and claim that those flaws prove that the videos are bad fakes. The irony, of course, is that if these really
were bad fakes they would probably look a lot more like the "no-planers" think real videos ought to!)