Stereo Photo-Metrology now a Sourceforge project

[Tduell]

Hullo All,
An announcement that might be of interest to some of you.
The stereo photo-metrology application that I talked about here
recently, has now been established as a Sourceforge project.
The project web site is at <http://stereo.sourceforge.net>
Have a look, and if you know anyone who might be interested, please
point them to/at it.
I need some help with development.

Cheers,
Terry

[luca vascon]

I'm dreaming a soft like that since Canoma years....
Do you think the process could be automated somehow, with CP aut finders like autopano or similar...
If we have a stereocouple with known stereo base etc??? and the object on rotating base...
I dream things like the 2 canon chdk hacked to stereoshot together.
The other dream would be to make it with a panorama stereo couple, maybe a 60 cm highness difference between them...
I could do it in a NICE place here in Venice for testing..
resolution will be the one you want!!!

2008/7/27 Tduell <tdu...@iinet.net.au>

[Yuval Levy]

Tduell wrote:
> I need some help with development.

<http://panospace.wordpress.com/2008/07/27/stereo-help-wanted/>

Yuv

[Tduell]

Huulo Luca,

On Jul 27, 10:02 pm, "luca vascon" <luca.vas...@gmail.com> wrote:
> I'm dreaming a soft like that since Canoma years....
> Do you think the process could be automated somehow, with CP aut finders
> like autopano or similar...

I realy don't know at this stage, but I suspect it should be possible.
With this sort of application, one first needs to have a set (a few)
known points in the scene, for which you know the x,y,z coordinates.
The datum can be arbitrary. These points need to be located within
each photo of a stereo pair, and there can be multiple pairs. It will
probably always be necessary to manually select these calibration
points, but once that has been done it may be possible for an
automated process to build a 3D model of a user selected region.
If one used a standardised calibration device set in the scene (say a
wire frame cube) it may then be possible for that to be detected and
the calibration to become automated. I'm not sure it is early days.
I would certainly like to some of the things that are taken for
granted in hugin, such as lens distortion correction, added into the
process.

Thanks for your comments.

Cheers,
Terry

[Tduell]

Hullo Yuv,

Wow. That is a nice story story.
Thanks.

Cheers,
Terry

[Tduell]

Hullo again, Luca,

On Jul 27, 10:02 pm, "luca vascon" <luca.vas...@gmail.com> wrote:

> I'm dreaming a soft like that since Canoma years....
> Do you think the process could be automated somehow, with CP aut finders
> like autopano or similar...
> If we have a stereocouple with known stereo base etc??? and the object on
> rotating base...

In my enthusiasm to prepare my previous reply, I omitted to mention
that this software doesn't require you to control the camera
positions. (If I have already explained this previously, please stop
me now!)
They can be arbitrary, handheld, with relative roll and pitch angles
between the images.
Stereo figures out the camera positions from the calibration data.
This simplies the picture taking, but does require a calibration
device, or suitably accurate measurements of points on an object in
the images. Obtaining x,y,z data is not easy, unless one uses the
corners of a building or somesuch.
To overcome this problem, I am thinking about a take-apart wire frame
cube, maybe a couple of sizes would suit various applications. Think
of some standard corner fittings, into which aluminium tubes are
fitted to assemble the cube, and it is levelled with some extensible
rods. the whole thing collapses into a small space and is carried in a
bag, a bit like hiking tent poles.
Assemble the cube, place it in the scene, level the cube, take a stero
pair, pack up, move on. When processing, choose one corner of the cube
as the datum, then all other visible corners are at known dimensions.
If two cubes are used, the first sets the datum, and the software
determines the relative coordinates of the second. for the next pair
of images, leave the second cube where it is, move the first (to now
become the third), the coordinates of the second cube can now be used
to calibrate this set of images and the coordinates will still be
relative to the first cube (datum), hence the surface model can be
indefinitely extended.
I have yet to test this, so I may come unstuck! let's say that it is a
desirable capability that may already be there.

Cheers,
Terry

[luca vascon]

You were perfectly clear...
You mentioned you can do it with every picture taken in the same scene.
I'd say it would be nice to be able to do things ALSO in that other way. It opens up to a lot of possibilities...
Did you ever seen Canoma?
It works the way you tell, to calibrate the camera.
Also Realviz has a similar software...

2008/7/28 Tduell <tdu...@iinet.net.au>

[Tom Sharpless]

Hi Terry,

I'll try to do a bit to further this project. I've always been
fascinated by 3D photos, and even find metrology somewhat
interesting. I know the art of what might be called "random view
photometrology" is now quite well advanced in the academic research
world; it's about time there was an open source exemplar.

I've been toying on and off with the idea of doing 3D with rotating
panoramic cameras, of which I have built a few. Such cameras are very
popular in robotics and reconaissance, because they are simpler to
calibrate than full-frame cameras, can give very high resolution at
reasonable cost, and generate data streams that are well matched to
the processing power of typical embedded computers.

It is a surprising fact that you can get stereo data from a single
point in a single scan with a single lens rotating camera. The trick
is to rotate about a point far from the NNP, and capture two slit
limages separated by a reasonable horizontal angle in the focal
plane. For details see, e.g. http://www.ima.umn.edu/preprints/mar2006/2105.pdf.
There is a better paper on this by a couple of Israelis, that I can't
locate at the moment.

I would have built such a camera already, but I don't think I'm up to
doing the optics necessary to project the image on two linescan
sensors (the ones I use are too wide to sit side by side in the focal
plane of a 35mm format lens). However a two-lens, two-slit camera,
which is way easier, would work just as well if not better.

One big advantage of using a stereo camera with a fixed geometry is
that you may not need any calibration device in the picture. You can
still take multiple pictures from multiple points of view, but each
view has a built-in calibration based on some near-to-middle distance
objects. You can bootstrap from these to a deeper 3D model of the
whole scene.

Regards, Tom

[Ken Warner]

Maybe this topic would make a good google group.

I'd like to lurk on such a group...

[Tduell]

On Jul 29, 3:32 am, Ken Warner <kwarner...@verizon.net> wrote:
> Maybe this topic would make a good google group.
>
> I'd like to lurk on such a group...

I suspect that at this stage there may not be enough lurkers ;-), and
certainly not enough happening just yet.
I had it in the back of my mind to set up such a group when there was
a bit of activity, but maybe there would be enough interested parties,
and it might stimulate some activity. There are some people who
frequent the GRASS GIS area that might be interested.
Thanks for suggestion, it has made me think a bit more about it.

Cheers,
Terry

[Tom Sharpless]

Hi Terry

There might not be enough workers, but I'll bet such a group could get
plenty of lurkers. Stereo photo analysis has many fascinating
ramifications.

One thing I'd like to explore is using stereo metrics to help deal
with parallax problems in pano stitching. In principle, if every
point of the pano appears in at least two images, and one has
disparity maps for all image pairs, then it is possible to interpolate
a parallax-free pano, such as would have been taken by a rotating slit
camera. Of course in practice there will usually be parts of the pano
that are only in one image, and extrapolating disparity maps into
those areas would probably not work well enough. But other approaches
to parallax might well benefit from some automatic disparity analysis
-- as might ghost reduction and even routine image alignment.

Ordinary pano stitching can be viewed as the problem of placing seam
lines, that tesselate the pano and outline the parts of the source
images that will appear in it. Artists from the Renaissance to today
have shown that compelling pictures can be put together from images
whose points of view are not all identical -- it is a matter of
putting the "seams" in the right places, and "blending" across them
convincingly. This is often called multi-point perspective.
I saw a rather amazing paper recently, demonstrating that this process
can be automated. It describes software that is able to emulate the
work of a San Francisco photographer, who creates enormously long,
very convincing, continuous multi-point panoramas out of photos taken
while walking the city's streets. Unfortunately I have mislaid the
reference, but will let you know when I find it. Anyhow, an analysis
of something similar to stereo disparity plays a big part in it.

Regards, Tom

[Tduell]

On Jul 30, 2:42 am, Tom Sharpless <TKSharpl...@gmail.com> wrote:
> Hi Terry
>
> There might not be enough workers, but I'll bet such a group could get
> plenty of lurkers.  Stereo photo analysis has many fascinating
> ramifications.

You are probably right. I guess we really should have such a group.

[your other comments noted]

Cheers,
Terry

[Yuval Levy]

Tduell wrote:
> You are probably right. I guess we really should have such a group.

correct me (everybody) if I am wrong, but I think that this list would
welcome such talk here - rather than having to subscribe to just another
group. and the synergies are more than apparent - CP, NPP, ...

Yuv

[Tduell]

On Jul 30, 9:17 am, Yuval Levy <goo...@levy.ch> wrote:

Hullo Yuv,

I have probably said already, that the more I think about Stereo, the
more I see similarities with Hugin/Panotools.
You have found the word that I had been looking for...synergy.
I suspect there will be a lot to be gained by developing Stereo with a
close eye on what Hugin has done/is doing.
The GIS area is the other primary user (that is apparant at the
moment) of this type of application, and their needs have to be
considered as well. It is essentially a GIS type role for which I
wanted Stereo.
I have had a project on the back burner for many years as a result of
there not being any cost or manpower effective method available to
me...until it dawned on me that stereo photos may be able to solve my
problem, and that sent me on a search which led to the Stereo software
package. That project is to document the hillclimb circuits/tracks
locally and (maybe eventually) nationally, and that grew out of my
work in simulation/modelling to predict hillclimb vehicle performance.
It seems that every thing is connected, somehow!
Anyway, the long and the short of it is that I am happy for discussion
to continue here. I have an interest in both topics.
As far as the GIS interests are concerned I could simply drop an email
to the GRASS GIS mailing list to say that discussion about Stereo will
be on this group, at least until it grows too big or diverges too far
away from this groups main interest.
I look forward to seeing more opinions on this.

Cheers,
Terry

[luca vascon]

I'd to say that the sinergy is true and deep. But we would better be very tidy with the post putting a <stereo> tag in the title. I'd say Hugin has got many other branches, touching any type of imagestacking software, raw developers, etc.

2008/7/30 Tduell <tdu...@iinet.net.au>

[Tom Sharpless]

Hi Terry et al.

I too would prefer to have Stereo discussions here for a while. If
there should ever emerge an active bunch of Stereo fans with obviously
divergent interests, that would be the time to start a new group.

-- Tom