[ptx] automatch: another attempt at semi-automatic panorama stitching

[Bruno Postle]

On Sun 04-Jan-2004 at 02:42:14 -0500, Ed Halley wrote:
> On Sun, 2004-01-04 at 14:29, Bruno Postle wrote:
> 
> > So I've written a proof-of-concept tool that implements my "angular
> > distance" technique of automatically detecting control points:
> 
> So this sounds like a good candidate for a feature launched on the
> Images tab of Hugin, when it gets to that stage of implementation.  The
> user would collect the images and then push an "auto-arrange" button on
> the Images tab or the Preview panel.

..or even point hugin to a directory full of photographs and let it
determine which of them can be assembled to form a panorama.

I suspect that this technique is quite sensitive to bad initial lens
parameters (fov, a, b, c), but I haven't had a chance to test it.

> Unless the images are fisheye, Hugin will have already gotten the
> HFOV and just needs to approximate the yaw/pitch/roll for each
> image.

(BTW hugin still detects all my photographs as 179 degrees fov)

> Does the method require any knowledge of the ultimate final projection
> (equirectangular, etc.)?  In other words, does projection choice affect
> the automatic discovery and selection of match control point features?

Not at all, the feature-points are mapped to a spherical surface in
3d space before being analysed, but this is necessary even if the
eventual result is a rectilinear or cylindrical panorama.

(That all sounds dreadfully complex, the math is actually very
simple and the "analysis" is crudely brute-forced)

-- 
Bruno