Aerial photographs mosaic II
515 views
Skip to first unread message
Demerval
Jun 20, 2011, 6:06:43 PM6/20/11
to PTGui Support
After accepting the suggestion of John Houghton in an earlier message
on this forum, compress some files and posted at the following address
to open this discussion:
http://oreades.org.br/admin/download/aerialphoto.zip
These are files that exemplify attempts to create a mosaic of 62
photos of one of the aerial surveys conducted in one region of Central
Brazil.
The photos were taken from an aircraft that played a flight plan
(tracklog in red points in overallmosaic.bmp) and the locations
specified were obtained from the pictures (green points in
overallmosaic.bmp).
In this example, the photos were obtained with overlapping 60% and 30%
(lengthwise and sideways, respectively). The four files DSC_13??.jpg
were obtained at the marked points in cyan (in overallmosaic.bmp) with
Nikon D300s equipped with a lens AF-S DX NIKKOR 18-55mm f/3.5-5.6G VR
(Set to 24 mm).
The PTGui is fabulous, you can quickly create mosaics very good.
However, after many attempts, have not found a way to minimize errors
that arise at the seams of the photos. Some were marked with red
circle in panorama.jpg (The circle on the right is what is the biggest
mistake). In photos of urban areas this problem is even worse.
I would like to know the best sequence of steps in PTGui which allows
the creation of a mosaic without this type of error, or a way to fix
them or minimize them.
Can I post more photos (or tif files with EXIF) if anyone interested
in making a more thorough test, but I think the four pictures
represent the problem well.
Thanks, John. I hope the group's help.
Regards.
Demerval A. Gonçalves
on this forum, compress some files and posted at the following address
to open this discussion:
http://oreades.org.br/admin/download/aerialphoto.zip
These are files that exemplify attempts to create a mosaic of 62
photos of one of the aerial surveys conducted in one region of Central
Brazil.
The photos were taken from an aircraft that played a flight plan
(tracklog in red points in overallmosaic.bmp) and the locations
specified were obtained from the pictures (green points in
overallmosaic.bmp).
In this example, the photos were obtained with overlapping 60% and 30%
(lengthwise and sideways, respectively). The four files DSC_13??.jpg
were obtained at the marked points in cyan (in overallmosaic.bmp) with
Nikon D300s equipped with a lens AF-S DX NIKKOR 18-55mm f/3.5-5.6G VR
(Set to 24 mm).
The PTGui is fabulous, you can quickly create mosaics very good.
However, after many attempts, have not found a way to minimize errors
that arise at the seams of the photos. Some were marked with red
circle in panorama.jpg (The circle on the right is what is the biggest
mistake). In photos of urban areas this problem is even worse.
I would like to know the best sequence of steps in PTGui which allows
the creation of a mosaic without this type of error, or a way to fix
them or minimize them.
Can I post more photos (or tif files with EXIF) if anyone interested
in making a more thorough test, but I think the four pictures
represent the problem well.
Thanks, John. I hope the group's help.
Regards.
Demerval A. Gonçalves
John Houghton
Jun 21, 2011, 3:53:22 AM6/21/11
to PTGui Support
Thank you for providing the sample photos. I managed to produce a
stitch that is good in some respects:
http://www.johnhpanos.com/aerial_pano.jpg (4MB)
This used viewpoint correction. There is some distortion in the image
shapes (i.e. not rectangles), and I don't think this approach would
work as well for the entire set of images.
I also tried Microsoft ICE, which offers three planar stitching
modes. That only gave a stitch without obvious stitching errors using
Planar 3, which produced similar distortions to those in my result.
Using Planar 1 mode, which allows only rotation, shifting and scaling,
produced similar stitching errors to those in your result.
It seams that alignment errors will be an inevitable consequence of
the subject not being flat (as would be expected). Stitching errors
may well be avoided or hidden at the expense of unacceptable image
distortions, though it's not clear how accurate you expect or need to
get.
John
stitch that is good in some respects:
http://www.johnhpanos.com/aerial_pano.jpg (4MB)
This used viewpoint correction. There is some distortion in the image
shapes (i.e. not rectangles), and I don't think this approach would
work as well for the entire set of images.
I also tried Microsoft ICE, which offers three planar stitching
modes. That only gave a stitch without obvious stitching errors using
Planar 3, which produced similar distortions to those in my result.
Using Planar 1 mode, which allows only rotation, shifting and scaling,
produced similar stitching errors to those in your result.
It seams that alignment errors will be an inevitable consequence of
the subject not being flat (as would be expected). Stitching errors
may well be avoided or hidden at the expense of unacceptable image
distortions, though it's not clear how accurate you expect or need to
get.
John
Willy Kaemena GM
Jun 21, 2011, 5:12:55 AM6/21/11
to pt...@googlegroups.com
As explained by Joost, PTgui was not intended to cope with this kind of problems, but there is software to solve this problem. Kolor is claiming to handle this problem, see here:
On Jun 21, 2011, at 10:50, PTGui Support wrote:
Hi Demerval (and anyone attempting aerial panoramas),
You should keep in mind that PTGui is not a tool for creating geospatially correct aerial images. It's designed to stitch photos taken from a single viewpoint, not from a moving airplane.
When the surface is not perfectly flat stitching errors are unavoidable because of parallax. Those stitching errors will not only cause visible misalignments, but will also cause mis-estimation of the lens parameters, resulting in curved images.
Also those stitching errors will accumulate over the entire image, causing images to 'creep' away from the correct geospatial location in the panorama.
Using viewpoint correction can be helpful if the images are not taken from the same elevation, or if the camera was not always pointing exactly downwards. But even viewpoint correction cannot eliminate parallax.
For the best results, control points should be placed in a single flat plane (i.e. only at sea level or street level). Everything in this plane will be stitched perfectly, lens parameters can be estimated correctly and the above creeping will not occur. Still misalignments will be visible above or below this plane. Since the automatic control point generator isn't able to detect if a point is above street level or not, this would require manual control point placement. Which is only practical if you are stitching a small number of images.
If geospatial correctness is important you might consider anchoring the panorama to an existing map or aerial image. Basically the process would be very similar to 5.33:
http://www.ptgui.com/support.html#5_33
But the above still holds; stitching errors will occur if the surface is not flat.
Joost
--
You received this message because you are subscribed to the Google Groups "PTGui" group.
To post to this group, send email to pt...@googlegroups.com
To unsubscribe from this group, send email to ptgui+un...@googlegroups.com
Please do not add attachments to your posts; instead you may upload files at
http://groups.google.com/group/ptgui/files
For more options, visit this group at http://groups.google.com/group/ptgui
Hugh
Jun 21, 2011, 5:48:15 AM6/21/11
to PTGui Support
Hello Demerval,
I have managed to slightly improve the fit by using the method
recommended by John and PTGui FAQ 5.6 and by adding additional Control
Points, but even adding quite a lot of new Control Points in the
problem areas do not eliminate the discrepancies.
The problem is that you are not really dealing with a "normal" case,
as John has pointed out.
The camera attitude (tips and tilts), difference in flying height and
differences in terrain elevation will all contribute distortions to
the images that PTGui is not designed to deal with.
Creating mosaics from aerial photography has always been a skilled
art, even using 230 x 230 mm format air survey cameras with surveyed
ground control points, so I think that what you are achieving with
PTGui is the best you can expect with this software.
PTGui is a great and very powerful software and, having spent some
four plus decades involved with photogrammetry, I really admire what
Joost has achieved, but as stated in FAQ 5.6 "PTGui was designed for
stitching panoramas, from photographs taken from a single camera
viewpoint. When stitching a panorama, images are warped to correct for
perspective distortion so that a seamless overlap is achieved."
The key phrase here is "from a single camera viewpoint", which is
quite different from flying aerial photography with forward and
lateral overlap.
A quick search or the Internet has not revealed any suitable solutions
so you may have to resort to getting help from people who have
experience in mapping from aerial photography using non-metric and
small format cameras.
There is a company based in England that has a lot of experience in
this field and it is probably worth your while contacting them for
advice or help.
They may even be able to suggest software designed for mosaicing
aerial photographs from small format cameras.
The company is Photarc.
www.photarc.com
I hope that this may be of some help.
Best regards, Hugh.
I have managed to slightly improve the fit by using the method
recommended by John and PTGui FAQ 5.6 and by adding additional Control
Points, but even adding quite a lot of new Control Points in the
problem areas do not eliminate the discrepancies.
The problem is that you are not really dealing with a "normal" case,
as John has pointed out.
The camera attitude (tips and tilts), difference in flying height and
differences in terrain elevation will all contribute distortions to
the images that PTGui is not designed to deal with.
Creating mosaics from aerial photography has always been a skilled
art, even using 230 x 230 mm format air survey cameras with surveyed
ground control points, so I think that what you are achieving with
PTGui is the best you can expect with this software.
PTGui is a great and very powerful software and, having spent some
four plus decades involved with photogrammetry, I really admire what
Joost has achieved, but as stated in FAQ 5.6 "PTGui was designed for
stitching panoramas, from photographs taken from a single camera
viewpoint. When stitching a panorama, images are warped to correct for
perspective distortion so that a seamless overlap is achieved."
The key phrase here is "from a single camera viewpoint", which is
quite different from flying aerial photography with forward and
lateral overlap.
A quick search or the Internet has not revealed any suitable solutions
so you may have to resort to getting help from people who have
experience in mapping from aerial photography using non-metric and
small format cameras.
There is a company based in England that has a lot of experience in
this field and it is probably worth your while contacting them for
advice or help.
They may even be able to suggest software designed for mosaicing
aerial photographs from small format cameras.
The company is Photarc.
www.photarc.com
I hope that this may be of some help.
Best regards, Hugh.
PTGui Support
Jun 21, 2011, 4:50:02 AM6/21/11
to pt...@googlegroups.com
Joost
>> Demerval A. Gon�alves
>
PTGui Support
Jun 21, 2011, 5:35:14 AM6/21/11
to pt...@googlegroups.com
Hi Willy,
That may be revolutionary for Kolor but what you see is just viewpoint
correction in action.
As far as I can see from the movie those images are taken from a
helicopter which was approximately at a static position so the parallax
is small. By contrast, Demerval's images are taken 100s of meters apart.
For an extreme example try visualizing the same castle but now with the
helicopter circling around it. It's easy to see that no stitching
approach would work: some images will contain only the back side of the
building, which is not visible in the panorama. The only thing you can
possibly do is align the photos at ground level, and discard any pixels
above ground level from all but one image. And that's exactly the
viewpoint correction approach.
Joost
>>>> Demerval A. Gon�alves
Willy Kaemena GM
Jun 21, 2011, 6:48:52 AM6/21/11
to pt...@googlegroups.com
Thank you Joost, as I said they are "claiming" it... I had no deeper look into it.
Willy
>>>>> Demerval A. Gonçalves
Demerval
Jun 21, 2011, 3:24:28 PM6/21/11
to PTGui Support
We use small format aerial photography (SFAP) as a tool to support
other remote sensing products, such as the validation of terrestrial
features observed in medium resolution satellite images.
They are not used necessarily for the extraction of metrics and their
quality ilustrative can be more important than their quality spatial.
We know that SFAP does not offer the quality cartographic equivalent
of those obtained from survey photogrammetry. However, the cost is
much lower, making them viable in certain applications.
We used the SFAP only individually through a catalog of georeferenced
points until we find rapid means of stitching.
Eliminate all mistakes is impossible, but I can improve a lot my
improve by following his tips:
- Use viewpoint correction* - FAQ 5.10 (John).
- Consider anchoring the scene to an existing map (Joost).
- Identify and test software suitable for multiple viewpoint -
Autopano, photarc suggestion (Willy and Hugh).
*I did a test with all the photos and it worked well. Apparently you
have to choose carefully the reference image. I don't know why
something goes wrong when setting the output file at maximum
resolution. I think I got the result I expected, but I will do more
tests.
Thanks to all.
Demerval
On Jun 21, 7:48 am, Willy Kaemena GM <wkaem...@googlemail.com> wrote:
> Thank you Joost, as I said they are "claiming" it... I had no deeper look into it.
>
> Willy
>
> On Jun 21, 2011, at 11:35, PTGui Support wrote:
>
>
>
>
>
>
>
> > Hi Willy,
>
> > That may be revolutionary for Kolor but what you see is just viewpoint correction in action.
>
> > As far as I can see from the movie those images are taken from a helicopter which was approximately at a static position so the parallax is small. By contrast, Demerval's images are taken 100s of meters apart.
>
> > For an extreme example try visualizing the same castle but now with the helicopter circling around it. It's easy to see that no stitching approach would work: some images will contain only the back side of the building, which is not visible in the panorama. The only thing you can possibly do is align the photos at ground level, and discard any pixels above ground level from all but one image. And that's exactly the viewpoint correction approach.
>
> > Joost
>
> > On 21-6-2011 11:12, Willy Kaemena GM wrote:
> >> As explained by Joost, PTgui was not intended to cope with this kind of
> >> problems, but there is software to solve this problem. Kolor is claiming
> >> to handle this problem, see here:
> >>http://www.kolor.com/blog-en/2011/06/10/palmela-2011-best-picks-how-t...
>
> >> <http://www.kolor.com/blog-en/2011/06/10/palmela-2011-best-picks-how-t...>Willy
> >>> For more options, visit this group athttp://groups.google.com/group/ptgui
other remote sensing products, such as the validation of terrestrial
features observed in medium resolution satellite images.
They are not used necessarily for the extraction of metrics and their
quality ilustrative can be more important than their quality spatial.
We know that SFAP does not offer the quality cartographic equivalent
of those obtained from survey photogrammetry. However, the cost is
much lower, making them viable in certain applications.
We used the SFAP only individually through a catalog of georeferenced
points until we find rapid means of stitching.
Eliminate all mistakes is impossible, but I can improve a lot my
improve by following his tips:
- Use viewpoint correction* - FAQ 5.10 (John).
- Consider anchoring the scene to an existing map (Joost).
- Identify and test software suitable for multiple viewpoint -
Autopano, photarc suggestion (Willy and Hugh).
*I did a test with all the photos and it worked well. Apparently you
have to choose carefully the reference image. I don't know why
something goes wrong when setting the output file at maximum
resolution. I think I got the result I expected, but I will do more
tests.
Thanks to all.
Demerval
On Jun 21, 7:48 am, Willy Kaemena GM <wkaem...@googlemail.com> wrote:
> Thank you Joost, as I said they are "claiming" it... I had no deeper look into it.
>
> Willy
>
> On Jun 21, 2011, at 11:35, PTGui Support wrote:
>
>
>
>
>
>
>
> > Hi Willy,
>
> > That may be revolutionary for Kolor but what you see is just viewpoint correction in action.
>
> > As far as I can see from the movie those images are taken from a helicopter which was approximately at a static position so the parallax is small. By contrast, Demerval's images are taken 100s of meters apart.
>
> > For an extreme example try visualizing the same castle but now with the helicopter circling around it. It's easy to see that no stitching approach would work: some images will contain only the back side of the building, which is not visible in the panorama. The only thing you can possibly do is align the photos at ground level, and discard any pixels above ground level from all but one image. And that's exactly the viewpoint correction approach.
>
> > Joost
>
> > On 21-6-2011 11:12, Willy Kaemena GM wrote:
> >> As explained by Joost, PTgui was not intended to cope with this kind of
> >> problems, but there is software to solve this problem. Kolor is claiming
> >> to handle this problem, see here:
>
> >> <http://www.kolor.com/blog-en/2011/06/10/palmela-2011-best-picks-how-t...>Willy
>
> >> --
> >> You received this message because you are subscribed to the Google
> >> Groups "PTGui" group.
> >> To post to this group, send email to pt...@googlegroups.com
> >> To unsubscribe from this group, send email to
> >> ptgui+un...@googlegroups.com
> >> Please do not add attachments to your posts; instead you may upload files at
> >>http://groups.google.com/group/ptgui/files
> >> For more options, visit this group athttp://groups.google.com/group/ptgui
> >> --
> >> You received this message because you are subscribed to the Google
> >> Groups "PTGui" group.
> >> To post to this group, send email to pt...@googlegroups.com
> >> To unsubscribe from this group, send email to
> >> ptgui+un...@googlegroups.com
> >> Please do not add attachments to your posts; instead you may upload files at
> >>http://groups.google.com/group/ptgui/files
CroHammeR
Jun 25, 2011, 2:56:33 AM6/25/11
to PTGui Support
I tested PTGui and Autopano for air panoramas (360 x 180 degree
spherical panorama). The images are very different and difficult to
stitch. The best result I got with PTGui.
Here are two examples:
http://www.vrpix.ch/flashpanoviewer/20110518_immobilie_riedsort_weggis.html
http://www.360cities.net/image/air-panorama-of-weggis-on-the-lake-switzerland
Reply all
Reply to author
Forward
0 new messages