In this tutorial - http://www.johnhpanos.com/starcal.htm - I describe how
to evaluate lens parameters by aligning camera images with a 360x180
equirectangular star image using PTGui, though Hugin will work equally
well. However, the method is somewhat flawed since due allowance needs to
be made for refraction in the earth's atmosphere, which changes the
apparent position of stars as you get nearer to the horizon. The method
does work quite well using the upper part of the visible hemisphere.
John
On Saturday, July 4, 2015 at 6:56:25 PM UTC+1, Benjamin Hill wrote:
> Super-resolution from highly overlapping photos seems hard. Maybe
> impossible?
>
> I have taken a series of night-sky images from a fixed camera over a
> period of several hours. I identified keypoints (stars) in the images, and
> correlated the keypoints across frames. This results in a file of ~1000
> rows of 4 ints:
> int FRAME_A, int STAR_B, int PIXEL_X, int PIXEL_Y (and timestamp of the
> frame if it would be helpful)
>
> The challenge is that it was taken with a wide-angle lens, and I don't
> know the exact distortion parameters of the lens.
>
> Goal: use the input file to either determine the lens model, or another
> mathematical model (machine learning?), that will allow a very precise
> function that maps a "true" night sky location to the coordinates in every
> frame that covers that part of the sky. It is ok if the model takes a long
> time to train, but the final lookup should be fast.
>
> getAllOverlappingExactSourceLocations(x1, y1) ->
> array of frames [{source_frame:frame_a, source_pixel_x:(float)x,
> source_pixel_y:(float)y}, {source_frame:...]
>
> This may be easier to think of in the inverse: given a panoramic landscape
> shot taken with a slowly moving motorized camera, and choosing any ray from
> the camera's location, get the intersections with that ray and all the
> photos that overlap that ray.
>
>
> It feels like this should be easy using Hugin or PTGUI: At the end of the
> day, I've got very accurate keypoints, and I *should* be able to get an
> accurate lens model from them... right? And once I have an accurate lens
> model, I should be able to get each frame's yaw/pitch/roll with sub-pixel
> accuracy. And then I... er... need help. How do I use this model + a
> frame's yaw/pitch/roll to get source pixel locations for the final frame,
> that is pretty much what the final stitcher is doing, true? So I'd need to
> reimplement my own emblend? Any easier way?
>
--
A list of frequently asked questions is available at:
http://wiki.panotools.org/Hugin_FAQ
---
You received this message because you are subscribed to the Google Groups
"hugin and other free panoramic software" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/hugin-ptx/406bb657-f2b7-42cb-95fd-8a89f1ddadae%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
