[ptx] panosifter problems & questions

[Terje Mathisen]

Bruno Postle wrote:

> On Tue 17-Feb-2004 at 10:37:00AM +0100, Terje Mathisen wrote:
> 
>>IMHO, the proper way would be to calculate all crop factors relative to 
>>a film/chip plane diagonal, since this diagonal determines the size of 
>>the optics, right?
> 
> 
> The annoying thing is that this would all be so simple if cameras
> just listed focal length in units of pixels - Everything else would
> be easy to derive whatever the orientation and aspect ratio.

Well, yes. :-)

Getting the diagonal right is still good enough to derive any other 
measurement in a line or two of code, so it really doesn't matter, IMHO.

What's slightly more interesting is the fact that due to 
barrel/pincussion distortion, a lens will have an effective FOV which 
depends on the amount of overlap between images, and not only the 
current focal length!

My Oly D40Z has an average case of distortion, about 1%, at the 
wide-angle end, but this is in the form of a gullwing, i.e. pincussion 
near the center and barrel at the edges.

What it really means is that when I make 360 panos, after getting all 
ypr values optimized, I then relax the hfov parameter and optimize 
again, this can sometimes lead to a very small modification of the hfov 
parameter which gives a significant overall improvement in the average 
control point error.

Terje

-- 
- <Terje.Mathisen at hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"