Mick is correct on the core error. MTS-B (standard on these IR feeds) maxes at 45° HFOV. Avi’s 65.6° best-fit exceeds that limit, so the 2.26 km altitude is invalid regardless of calibration score. Mick’s ~1.5 km / lower FOV aligns with hardware and visual Sitrec fits. Point clustering makes the high solution unstable. Pixel rate is solid; the absolute scale is not.
Excellent commitment by @JoelValdezDOW! This is the kind of response we should strive for when new UAP information emerges. I support this effort 🫡 🇺🇸 https://t.co/3rIAUmBSU2
There is nothing wrong with a standard planar camera pose reconstruction, as used by Nandal and Loeb. (I use a more accurate 3D solver, but the results are similar here) The problem is with the input data. They used a tight cluster of points that were poorly conditioned to resolve to a camera position. So they got inaccurate results. GIGO. Doing lots of work is great. It’s just not a guarantee of accuracy. Here the video is ground truth. Nandal’s altitude does not match the video, so it’s wrong. The Sitrec derived altitude is a perfect match, so its correct. We need to test the numbers against the ground truth. I’ve done that.
To appreciate how much more work was invested in our calculation compared to Mick’s superficial tweets, please check the full report. We really break down what we do with actual numbers. Our 2.26 km altitude comes from a standard planar camera pose reconstruction, with the site and terrain geometry independently checked in Google Earth and Google Maps. We systematically swept the unknown HFOV. The calibration score reaches its minimum at 65.6 degrees, giving 2.26 km, with the 5 percent near-minimum range spanning 60.4 to 71.0 degrees. The same solver reproduces your quoted 1.465 km at about 34.5 degrees. However, its calibration score is 0.744 compared with 0.308 for our preferred solution. Because lower scores indicate a better fit, the 1.465 km solution scores about 2.4 times worse than our preferred 2.26 km solution. Separately, we ran 3000 Monte Carlo perturbations of the image and map control points. These give a ground-plane speed of 3.37 km/s, with a 68 percent interval of 2.88 to 4.02 km/s. Most importantly, the measured motion does not depend on choosing 1.465 or 2.26 km. The object moves at about 801 px/s over 0.4004 s relative to the stabilized terrain. Changing the assumed HFOV does not change that measurement. Nor does the 2.26 km altitude produce our 3.37 km/s ground-plane result. It only affects the later conversion to a 3D physical speed for an object at an assumed altitude. The full derivation and uncertainty analysis are given in our paper. https://t.co/MN9WLLe4ia
375 source assets indexed · Release 05