(meteorobs) Looking for decent references on Fisheye to Az/El methods
falcon99 at sbcglobal.net
Sat Aug 27 16:49:38 EDT 2011
I've seen a few posts here and there eluding to this, and am sure there is a LOT of data out there somewhere - the ASGARD system is a demonstration of the success, as an example.
I am looking for decent methodology references to convert from m x n pixel locations (x,y) to angle from camera boresight for a fisheye-type lens CCD image, given known first or second order parameters such as focal length, and pixel values such as image boundaries and center.
Now, this may sound pretty simple, and one would expect to find lots of readily available references, but I'm coming up short. Understand that I'm not an optics or math guy, but I given good papers and texts, I can work a good method (I am an engineer by trade).
Most papers and optical processing methods I've found fall into the following categories:
1. Just assume some kind of translation is done, but no details given to the method. e.g. We start with pixels, then after rotating coordinate systems from boresight to X,Y,Z, then something magically occurs to translate the fish-eye image aberrations. Not sure what the authors are doing here.
2. A general assumption regarding pixel to cone angle translation is made that is extremely general and, in most cases, grossly incorrect. One example was Theta = 2 * asin (R/2f) where f = focal length in mm and R is the center to pixel distance in pixel units. I have no idea what the author was trying to convey, but the lack of unit translation was disappointing. Some of these are dissertations. Had I gone for a master's degree and done this, my profs would have had me for an afternoon snack.
3. The author gave up on any modeling methods and hand calculated residuals into a look-up table and called it good. (Most likely what I'll end up doing, but would have to repeat it every time I clean or refocus the camera).
So, knowing that there may be some lurkers on the list who have probably suffered through the same confusion, can any of you point me to a more simpler, and hopefully accurate source of methodology.
The goal on my end is to come up with a rough translation from Sandia Allsky images / movies to a 4-D (n, az, el, time) dataset. (n for the advanced movie case that can be used to produce angular position and velocity values).
This isn't for anyone in particular, just my own hobby-level curiosity. The end-goal is a matlab procedure taking two images or movies, doing pixel-level calculations, translating those into ECEF or E-N-U coordinate systems, and ultimately writing a KML or other graphic-friendly output for 3-D approximation. KML is easy. Matlab methods shouldn't be too hard. But the math to get through the camera translation is getting painful. Azimuth is a no-brainer.
The biggest unknown is what to do with the fisheye lens off-boresight translation.
Anyone have any good references?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the meteorobs