Using Leica DMC III images

[radar_boy]

I would like to use some stereo, vertical Leica DMC I and DMC III aerial imagery, provided as RGB in TIFF format. I have exterior orientation metadata as text, together with some interior orientation metadata (e.g. calibrated focal length, pixel size etc). I can import said data and metadata.

I have started by working with a block of >70 DMC III images and Align Photos to create a sparse cloud.

Unless I deliberately fix the focal length it ends up changing dramatically (>34000 mm vs. the calibrated 92mm). When I then use Model > Gradual Selection and progressively select points in the sparse cloud according to Reconstruction Uncertainty I get distinct structures forming:





When I fix the focal length the sparse point cloud appears thus:



These structures suggest systematic errors. I sense there is something wrong with the camera model? Is there something extra I should be doing to import Leica DMC images? Might the systematic errors be related to the fact DMC use multiple 2D arrays?

[Paulo]

Hello radar boy,

coud you show the Camera Calibration Inititial pane for your DMC III project?

The value for f should be as shown in following screen capture in case of a camera with following camera calibrated report...

Code: [Select]

Geometric Calibration
The output image geometry is based on the Pan Camera head (reference head =
master camera). All other camera heads are registered and aligned to this head.
Aerial triangulation checks overall system performance based on.
Output image
Reference Camera PAN
Serial Number 00128301
Number of rows/columns [pixels] 25728 x 14592
Pixel Size [µm] 3.900 x 3.900
Image Size [mm] 100.3392 x 56.9088
Focal Length [mm] 92.0000 mm + /- 0.001 mm
Principal Point [mm] X= 0.0000 mm, Y= 0.0000 mm + /- 0.001 mm
The “SYNTHETIC” geometric calibration is based on a simulated mathematical lens
distortion calculation based on the detailed optical design data of the lensas these high level cameras are synthetically calibrated all values should be fixed.....

as for the reconstruction uncertainty level threshold, a value of 10 corresponds to a good geometric reconstruction, so I would not go lower than 10 in threshold selection....

For these high resolution images (more than 350 Mpx), I would recommend using Guided image matching option in Align Photos... and maybe 500 for Key point limit per Mpx depending on your computer power .....

[radar_boy]

Many thanks for your reply Paul 

I have a PDF which contains the info in the same format as you presented. For anonymity I've removed the Serial Number, but everything else is just a copy & paste:

Code: [Select]

Geometric Calibration
The output image geometry is based on the Pan Camera head (reference head =
master camera). All other camera heads are registered and aligned to this head.
Aerial triangulation checks overall system performance based on.
Output image
Reference Camera PAN
Serial Number NNNNNNNN
Number of rows/columns [pixels] 25728 x 14592
Pixel Size [um] 3.900 x 3.900
Image Size [mm] 100.3392 x 56.9088
Focal Length [mm] 92.0000 mm + /- 0.002 mm
Principal Point [mm] X= 0.0000 mm, Y= 0.0000 mm + /- 0.002 mm
The “SYNTHETIC” geometric calibration is based on a simulated mathematical lens distortion calculation based on the detailed optical design data of the lens.

[radar_boy]

Here is my Camera Calibration GUI. I believe the only difference with yours is I have fixed f

[Paulo]

Yes, radarI

I woud fix all parameters.....

[jrp]

Hi, I don't understand the details of what's being discussed here, but I do recall that agisoft sometimes refers to image parameters in pixels, not in mm.

Is it possible that 34000 pixels is actually a correct focal length in this context?

I've never used anything like that model of camera, so I have no idea what's inside it.

[radar_boy]

I woud fix all parameters.....

I'm returning to this processing again and just want to triple check that the focal length, pixel sizes and principal point coordinates are fixed to those from the [synthetic] camera calibration, and all other values are also fixed, presumably at zero?

If so, once erroneous sparse points are removed and Cameras Optimised (i.e. final bundle adjustment) only the exterior orientation (i.e. camera stations/poses) are adjusted? If this is correct, I'm curious to know what kind of Accuracy you used for the camera locations (m) and orientations (degrees) please? Did you import values from a table of ephemeris data for the flight (I have SD for locations, via dGNSS, and orientations, via IMU), or did you set Accuracy values the same for all photos? If the latter is the case, did you use small/tight values (e.g. fractions of a metre and fractions of a degree) or larger/looser values (e.g. a metre or more and a degree or more)?

Many thanks again! 

[Paulo]

Hi radar boy,

I am not using this system per say. but for examaple flying with post processed Applanix GNSS/IMU POS sytem you would get something like 10 cm accuracy or better depending on metho dused for pos processing on camera centers and something like 0.01/.03 for accuracy of orientation... For the system you are using, what POS system used and the accuracy obtained....

Cheers,