Some nice big mosaics

[Tom Sharpless]

I've been happily making high resolution mosaics of wall paintings
with Hugin. There is a page of them at
http://tksharpless.net/PhillyMurals/.

They were shot with 15mm, 24mm, and 35mm lenses on APS-C format
cameras, from multiple positions about equidistant from the wall,
using a pole to get a good high point of view. In almost all cases
there were poles and wires in front of the pictures. I have been able
to 'vanish' most of them just by masking before blending, but in a few
cases I have resorted to hand painting in Photoshop.

It's a bit chancy getting the very wide ones to align right, but with
plenty of straight line CPs and care to avoid CPs not on the plane of
the mural, I've had good results. In some cases I've had to add one
image at a time to the optimized pool, presetting y,p,r,x,y,z to
reasonably close values.

Anyhow I'm quite pleased with Hugin's ability to do this, which PTGui
cannot.

Cheers, Tom

[kfj]

On 12 Jan., 18:48, Tom Sharpless <tksharpl...@gmail.com> wrote:
> I've been happily making high resolution mosaics of wall paintings
> with Hugin.
>

> It's a bit chancy getting the very wide ones to align right, but with
> plenty of straight line CPs and care to avoid CPs not on the plane of
> the mural, I've had good results. In some cases I've had to add one
> image at a time to the optimized pool, presetting y,p,r,x,y,z to
> reasonably close values.

I've done a fair amount of similar stitches (mainly maps and
signposts), and I found that trying to optimize y,p,r,X,Y and Z all at
once introduces to many degrees of freedom, so the whole thing totally
fails. But you can still work on the whole image set at once if you
only only do a few parameters at a time. I often start out with just X
and Y and then take it from there, optimizing the others in turn. Once
everything has 'gelled' nicely, you can optimize more parameters
simultanously. Doing it this way is probably faster than introducing
the images one by one and estimating parameters. The process also
benefits from lots of CPs (like, 100-200 per pair).

This makes me wonder if there couldn't be a specific mosaic
optimization mode which follows such a pattern, thus avoiding the mess
that usually happens if you try optimizing everything at once.

Kay

[Tom Sharpless]

Hi Kay

In my experience the most important things are to place lots of
straight line CPs, and to rigorously exclude both off-plane and bogus
CPs. Hugins CPFind can generate a lot of false matches when matching
'all vs all', so I try to organize my shots so that 'multirow' mode
works. But in any case I check every CP visually.

Straight line CPs are the 'backbone' that keeps the whole pano
square. They only work when the target projection is rectilinear, and
at first the optimizer wants to generate yaw angles that are too big.
So I have to reset and reoptimize from scratch a lot -- just positions
and viewpoint, with nominal lens parameters. Eventually when the CP
set is good, the pano snaps into alignment and I can then tweak lens
parameters if it seems necessary -- though with a calibrated lens it
rarely does.

Regards, Tom

[Matthew Petroff]

I agree that straight line CPs are extremely important. When I
stitched a large mosaic (135 images with an 8mm fisheye, ~1800 manual
CPs), I used a pre-calibrated lens and optimized position and
translation after every image [1]. At first, I started off without
using straight line CPs, and it did not turn out well; adding straight
line CPs, however, did wonders. After, I still had to do a lot of
tweaking and masking to get better alignment, but the straight line
CPs were crucial in maintaining the overall shape of the mosaic.

Matthew

[1] http://www.mpetroff.net/files/openzoom/taft-class-of-2011/