Leonardo wrote:

Hi all,
I am a new Autopano user and I've started a new big project to create a Giga photo. I am trying to do an ortophoto (so I need a planar projection) from a panorama made by 224 pictures. All the pictures are taken with a GigaPanEpicPro, and I've used a D800 with a tele 200mm focal lenght (this is why I needed 224 pictures to cover a wall of about 12mX10m). The final RMS of the stitching is very good (1.95).
My problems is that I can't correct the strong deformation in th high part of the panorama (marked in red in my snapshot). I've tried to use vertical and horizontal correction and also the transformation tools (yaw, pitch and roll) without any result. I am sure that the wall is almost a perfect rectangle, so my questions are:
Why do I have these kind of deformation?
How can I adjust them?
Thanks in advance
Leonardo

Hmm, maybe i am wrong, but try to use the "move/drag complete pano" to move the complete panorama a little bit up, down, left, right.

Getting better?

greetings from germany
Chris

Christian Stüben wrote:

Leonardo wrote:

Hi all,
I am a new Autopano user and I've started a new big project to create a Giga photo. I am trying to do an ortophoto (so I need a planar projection) from a panorama made by 224 pictures. All the pictures are taken with a GigaPanEpicPro, and I've used a D800 with a tele 200mm focal lenght (this is why I needed 224 pictures to cover a wall of about 12mX10m). The final RMS of the stitching is very good (1.95).
My problems is that I can't correct the strong deformation in th high part of the panorama (marked in red in my snapshot). I've tried to use vertical and horizontal correction and also the transformation tools (yaw, pitch and roll) without any result. I am sure that the wall is almost a perfect rectangle, so my questions are:
Why do I have these kind of deformation?
How can I adjust them?
Thanks in advance
Leonardo

Hmm, maybe i am wrong, but try to use the "move/drag complete pano" to move the complete panorama a little bit up, down, left, right.

Getting better?

greetings from germany
Chris

Hi Chris,
thanks for your answer. I've tried with the movement of panorama, but it didn't change the curvature above. And it is very difficult to move the panorama because for each little movement of the mouse corresponds a huge movement of the pano. Is there a more precise way to do that, even if I think it will not correct my deformation?
greetings from Italy!

Leonardo wrote:

And it is very difficult to move the panorama because for each little movement of the mouse corresponds a huge movement of the pano. Is there a more precise way to do that, even if I think it will not correct my deformation?
greetings from Italy!

You are working on a Laptop (wild guess). The mobile video chip on my laptop has this weak response behavior too. Better and bigger video cards are reported to have much better response times.

If not yet done, try to activate GPU for the edit window.

greetings from germany
Chris

Last edited by Christian Stüben (2013-02-14 19:35:55)

Leonardo wrote:

But in Autopano I choose a planar projection (because I need an orthophoto) and in this way all real straight lines should be projected in straight lines. So my problem is why in the top of the panorama, the real horizontal lines are projected in a kind of parabolic lines.
Let me know what you think
Bye
Leonardo

It seems to me that what Kolor usually calls a 'planar projection' is what is elsewhere termed a 'rectilinear projection' - in fact the Kolor documentation about Projection modes also (alternatively) calls it rectilinear. The 'deformation' you see is an inherent characteristic of a rectilinear projection:

http://www.autopano.net/wiki-en/action/ … ting_Modes

http://wiki.panotools.org/Rectilinear_Projection

What you seek is somethingelse altogether.

Last edited by mediavets (2013-02-14 21:09:42)

lumelix wrote:

Hi Leonardo
For creating real orthophotos you need measured control points and a special software.
It's because the software must know the distances to the points in the picture.
Have you considered the nodal point of your camera/lens-combination ?
For creating orthophotos, some experience isn't wrong.

The planar projection in APG has nothing to do with this, it's only a planar surface where the image
is projected to, but it's not this wall

Hi all,
thanks for all your replies. I was reading this article in mathworld:
http://mathworld.wolfram.com/GnomonicProjection.html
and it seems to me that Gnomonic Projection is exactly what APG calls planar projection. But if this is right, you can see in mathworld's link, that just great circles are projected in straight lines. The curved line in my panorama is very similar to that ones distant from the equator in gnomonic projection (It seems a conical line).Again, if this is right, why APG says that real straight lines are projected in a plane again in straight lines?
For the orthophoto I would like to obtain, I know that in APG I miss the real dimension, but I was thinking: if I make a rectilinear panorama without distortion, then I could scale it respect the dimension I took with a 3D disto, so to obtain a scaled representation of the wall; do you think this is correct?
Thanks again
Leonardo

Ok Andrew!
so if the FOV is low I have real rectilenear  line that are projected in rectilinear line in the plane. In my pano probably the FOV is very high and so lines far from the principal points are curved.
At this point what do you think I can do? Probably the solution is to take again the panorama from a bigger distance (to reduce the FOV), but I can't do this because there is a huge chandelier which will cover the wall.
Any other solution?
Leonardo
P.S. I didn't say before, but I took the pictures rotating around nodal point. By the way I've written some articles about the battle between nodal point vs. pupil point: where is the truth?

mediavets wrote:

Leonardo wrote:

Again, if this is right, why APG says that real straight lines are projected in a plane again in straight lines?

Leonardo

You missed the proviso 'if the angle is low':

http://www.autopano.net/wiki-en/action/ … projection

lumelix wrote:

The distortions would be better if you shoot the wall from a platform in the medium height.

Yes - definitely. The problem here is the distance - this would be compensated by a higher point of view . . for example a
tripod which you can extend to 3 - 4 meters height.
(a platform you´re most unlikely allowed to take into such places . . )

I use a Gitzo tripod which i can lift up to about 3m - you´d be surprisd how helpful it is on many occasions in terms of getting
a more  suitable perspective!
(and the guards in a museum or in a church don´t feel they need to push the panic-button the moment when you enter the location with
hazardously looking equipment . . . - but good to inform the director anyway before you appear there . . )

Ideal for such situations is the big lamp-tripod from Manfrotto or Avenger which you can rise up to  4 or 5 m having a remote-controllable camera/head
mounted on the top.

best, Klaus

Last edited by klausesser (2013-02-15 16:45:35)

Leonardo wrote:

So my problem is why in the top of the panorama, the real horizontal lines are projected in a kind of parabolic lines.

Because you shot a too big FOV on a too short distance using a too long lens for this purpose
May i ask you: did you try my suggestion using the vert./hor. line tool? At lest i would give it a try.

best, Klaus

Last edited by klausesser (2013-02-15 16:53:48)

Leonardo wrote:

Hi Klaus, and Hi all,
yes I've tried with horizontal and vertical lines but without success. I was thinking about gnomonic projection, and I made a geometrical model of my wall. I was sure that curved lines in a gnomonic projection should depend by the distance from the principal point (named Oo). But take a look to my image:
if I project in a plane ABCD tangent in Oo to the sphere from its centre O, the real straight line AD remain straight after its projection (see the view from nodal point). But if i project the parallel circles (in blue) they will be transformed in straight line just for the equator circle and became conical curves for the lines going to the Poles. So, in an ideal projection model, line AD should be again a straight line! Ideally AD could be far away but it would be projected in a straight line again.
So I am again at the start point: why my A'D' line of my pano is curved? It dosen't depend by the model of projection but probably by optical distortion. I think my geometrical model is correct but let me know what do you think.
Thanks again for your patience!
Leonardo

We photographers call it: perspective.

best, Klaus

Leonardo wrote:

Hi Klaus, and Hi all,
yes I've tried with horizontal and vertical lines but without success. I was thinking about gnomonic projection, and I made a geometrical model of my wall. I was sure that curved lines in a gnomonic projection should depend by the distance from the principal point (named Oo). But take a look to my image:
if I project in a plane ABCD tangent in Oo to the sphere from its centre O, the real straight line AD remain straight after its projection (see the view from nodal point). But if i project the parallel circles (in blue) they will be transformed in straight line just for the equator circle and became conical curves for the lines going to the Poles. So, in an ideal projection model, line AD should be again a straight line! Ideally AD could be far away but it would be projected in a straight line again.
So I am again at the start point: why my A'D' line of my pano is curved? It dosen't depend by the model of projection but probably by optical distortion. I think my geometrical model is correct but let me know what do you think.
Thanks again for your patience!
Leonardo

Did you ever use a large-format-camera?

Ok: the edvantage of a large-format camera is - among others - that you can rise and/or fall the lens-board and the film-/sensor-boear independently.

This way in your case the procedere would be:

1) level the camera.
2) rise the lens-board parallel (!) to your wall. We call it "shift"
3) Shoot from the widest distance you can go.

Given you have a wide-enough lens it will cover what you need. (i´m still talking about large-format photography).

Very important is the distance between lens and wall. If you´re standing too close you either wouldn´t cover your desired FOV or you´ll get extreme (!) distortions.

Your horizontal as well as your vertical lines wil be straight as far as it´s possible - because your lens projects the part of a sphere symmetrically onto a plane: the film-surface.

The point is: your camera MUST be levelled perfectly. Your lens-board MUST shift perfectly parallel to the wall.

The problem is: everything which is not the on the wall-plane but stands in some distance between you and the wall will be extremely distortet.

The solution shooting it sequentially (stitch):

1) rise the camera to a higher point.
2) level the camera perfectly.
3) shoot symmetrical rows starting from the horizontal row - which should represent the levelled row.
4) shoot a MUCH bigger FOV than you need in the end.

I often use sequential shooting as a substitute for a shift/tilt camera. It works - if you shoot symmetrical rows.

When you shoot symmetrical - same angle and amount or rows up and down - you will be able to do a symmerical stitch. Without curved horizontals.

best and good luck, Klaus

Leonardo wrote:

So do you think my problem is in the gap beetwen ideal geometry and real geometry of photography?

Hi Leonardo
Exactly! Your graphic shows it very nice. In theory you use a pinhole camera in (O). If you shoot this wall with a pinhole camera, you get exactly straight lines all over the image.
But you don't. I suspect that you are shooting with the AF-S 70-200mm ? It's a very complex zoom lens, build with 21 elements and with a very complex optic. Even the very sharp AF Micro Nikkor 200mm has 13 elements inside and also a very complex optic
(See this: http://www.pierretoscani.com/echo_telez … glish.html).

It's very difficult to find the exact nodal point for the normal FOV of these lenses (ca. 12º) at a defined focus distance. But your FOV is about 12 times larger. And perhaps you have change the focus distance for the upper images ? So you change the nodal point as well. And the lenses create a distortion in every image. And then, APG try to find a good solution for the stitching and all these parameters shown in the list.

Into aerial photogrammetrie were we would create maps from orthophotos, shooted from a airplane, we use so called calibrated cameras and high precision lenses (eg. from ZI, Leica or Zeiss). For the distortion correction of these images, we used messured controlpoints and special software.
Also in the close-range photogrammetry we use calibrated special cameras. But actually, most of projects are made with modern laser scanners. They can measure the distance to the points and take pictures at the same time, so you get a fully 3D-modell.

If you don't need this high resolution, it would be easier to shoot with a shorter focal lens. Use a 55mm Micro Nikkor or a PC-Nikkor. Set the nodal point as precisely as you can and the shooting position as high as possible. Do not change the focus distance.
Another possibility could be the stereophotogrammetry. The wall would be shoot from two positions. Then a special 3D-software can calculate the exact dimensions. A surveying company would use for this work today a 3D laser scanner.

Leonardo wrote:

I've used a Nikon tele 200m with a fixed focus so I didn't change the nodal point during the shooting, but at this point I'm not sure to have found the nodal point: I think it's impossible! Thanks for your links. I think we will take some pictures to make a photomodeling of the wall, to be sure of its geometrical shape.
What software do you use in photogrammetry?
I think next time I will use 105mm

Shooting indoors with 200mm means to absolutely need an EXTREMELY precise setting of the NPP. 200mm are VERY long related to the camera-/object distance.
Shooting indoors a very wide angle using a 200mm lens means you NEED to shoot symmetrically - the same + and - rows for being able to do a proper planar-projection.

Your shoot was non-symmetrical - that means: you would have needed to use the same angle for + as for - . Simply said: when you shoot 75° up you need to shoot 75° down also for being able to do perfect un-distortion. The same is with left and right btw.
Then you stitch it ALL - and run the geometrical corections. After rendering you can delete the redundant parts of the image.

The best way to virtually visualize it is to think about how photography works . .   Doing ONE shot you would have get ot done by using a shift-lens: camera perfectly levelled so that the lens-axis is parallel to the ground and the lens shifts parallel to the wall. But that´s impossible when you use a 200mm lens.

Using a shift-lens for a stitch-shot definitely is useless.

best, Klaus