[GRASS-user] Orthorectification of tilted digital camera images

[Blumentrath, Stefan]

Dear all,

Although there is a lot of relevant information regarding the orthorectification of tilted digital camera images in GRASS GIS online (here:
http://grasswiki.osgeo.org/wiki/Orthorectification_digital_camera
http://www.academia.edu/853401/Robust_rectification_of_aerial_photographs_in_an_open_source_environment
http://osgeo-org.1560.x6.nabble.com/rectifying-digital-images-td3929803.html
http://www.grassbook.org/wp-content/uploads/neteler/papers/neteler2005_IJG_051-061_draft.pdf
and the module manuals of course), I am still struggling with applying the procedure to my example properly. The problem is that the image is either not enough or too much stretched and twisted and never matches with the GCPs...

I have an image taken with a wildlife camera trap which shows four color markers we placed in the terrain and measured using a handheld GPS (aiming at using those as GCPs). Camera position has been GPS-measured as well. I do have a high resolution DEM (Lidar) for the area and enquired camera specs from the manufacturer (Reconyx) as the camera does not produce EXIF data...

The camera has a fixed focal length of 8.5 mm, a sensor size of 4.8 x 6.4 mm with 1536 x 2048 pixels.
So I defined 4 fiducial marks (starting clockwise from middle of the left border) as
In mm on the sensor
1 -3.2 0
2 0 2.4
3 3.2 0
4 0 -2.4
Corresponding pixels on the image are defined as
1 0 768
2 1024 1536
3 2048 768
4 1024 0

RMS is 0.0

When it comes to defining initial camera exposure I get quite a bit unsure about my settings (and the camera exposure settings seem to have a significant effect on the results).
X,Y,Z should be OK, as they are measured by GPS/DEM OK.. The article here http://www.grassbook.org/wp-content/uploads/neteler/papers/neteler2005_IJG_051-061_draft.pdf says roll, pitch and yaw can be estimated using d.measure.
The photo direction ("yaw") should be ca. 270 degree as the picture is taken in towards east direction. But how do I estimate "pitch" and "roll"?
Given that the picture is taken slightly uphill (and even shows some sky) I tried 120 degree for "pitch". But that worked equally bad as e.g. 90 or less... (I suspect "pitch" to be the main source of error)
I kept "roll" more or less constant at 0 as it was hard to guess.
Does up-slope direction of the photo or sky in the photo conflict with the orthorectification algorithm?

When it comes to the GCPs I am in principle quite confident that they are placed OK. Or do you think that GCPs measured by a hand held GPS are too imprecise?
However, depending on initial camera settings RMS something like 268, which indicates that something is wrong here. If I do not use i.photo.init, the image is placed much better (RMS around 10-20), but still pretty poor...

I can provide the data if that helps and if someone would be willing to take a look at them after reading my long mail (sorry).

Thanks for helping in advance,
Stefan

Btw: Some of the submodules of i.ortho.photo are not linked to the /bin folder in GRASS 6.4.5svn, so I have to start them like this: /usr/local/grass-6.4.5svn/etc/i.photo.init

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20150811/5ff7e22c/attachment-0001.html>