Help removing the "Bowl Effect"

[cfc18]

Hello,

I am working on making a DEM out of 20 images from the NAIP but have gotten some variation of the bowl effect a number of times now. I’ve read through the forums and tried the various solutions there, but nothing I read has solved this issue.

It seems like this is typically an optimization problem. I have tried optimization with all fits, all fits except k1/k2/k3, all fits except p3/p4/k4, and all fits except p3/p4/k3/k4. Optimization with all of the fits and with all except p3/p4/k3/k4 gave me a dome, while all fits except k1/k2/k3 and all fits except p3/p4/k4 put square shapes on the DEM where the overlapping images were.

For reference, I should have the correct coordinates for the cameras and GCPs, I have camera accuracy high (~200m, unchecked) and a reasonable GCP accuracy (~5m, checked, the images I’m working with are pretty detailed), and have 31 GCPs. The images were taken on the same day by a CAM5367, which has a focal length of 153.814 that I input into the system. CAM5367 has no values for fits p3/p4/k3/k4, so I unchecked them in the next optimization and still got a bow-shaped DEM.

If anyone knows how I can solve this problem, it would be a huge help.

[Paulo]

Hey cfc18,

i suppose this is scanned film photography....( 23 cm by 23 cm film format with 150 mm focal length)

Could you show the initial camera calibration window to see if initial values make sense?

or maybe share a link to imagery so we can try on our side?

One other topic is camera fiducials in scanned photos.... you should have them defined and detected according to Referencing chapter 4 of help menu contents. See attached screen capture

[cfc18]

Hi Paulo,

Yes, the dimensions are approximately 24cm by 24cm with a focal length of 153mm.

I have the initial camera calibration attached below. I input the focal length manually (when the pictures were first uploaded, if was “unknown”) and then ran optimization, but otherwise, all of the values are the same.

I’m not sure if I’m allowed to send the imagery. I’ve attached a screenshot of one of the NAIP photos with the fiduciary. If you need more than that, let me know, and I will check with my research professor. There may be a problem with the focal length of the images. According to the camera type, the focal length is 153.814mm, but under properties it says the focal length is 35mm. Could this be the issue? I’ve ran optimization under both focal length and it came back bowl-shaped both times.

Regarding the fiduciary, it was recommended to me that I remove them, so I batch trimmed the fiduciary off of all of the photos in Photoshop. This should have also ensured that the images were all the same size relative to the center. Then, I converted the coordinates to easting and northing to match the geoid I am using and converted the drone height from feet to meters, then imported these coordinates into the system for camera coordinates. In short, the fiduciary was taken off, so the fiducial marks were not referenced.

Thanks,
Camille

[Matt]

You are better off not using your predefined camera focal length and letting photoscan work it out for you during alignment.
Mask the edges of the images and off you go.  I have successfully applied this methodology to scanned photos from the 1930's and with good ground control you can georeference the cloud and produce an orthomosaic to within a pixel or two.

[cfc18]

Hi Matt,

Thank you for your reply! In previous models, I didn’t adjust the focal length and left it as “unknown” for optimization. My DEMs for these models were bowl-shaped as well.  In total, I have tried three focal lengths (unknown, 35mm, and 153.8mm) with varying optimization.

I was told that trimming the fiduciary off of the images and then importing the images had basically the same effect as masking them. I’ve been georeferencing my GCPs in ArcMap, so those should be pretty accurate. I think our only difference in workflow is that I went straight from the dense cloud to creating the DEM while you produced an orthomosaic, and I don’t think that alone would produce the weird elevations I’m getting.

[Paulo]

Definitely, the focal is 153.814 mm as this is a Wild classic wide angle lens commonly used in photogrammetry....

from the scanning resolution 16 712 pixel, it seems the pixel size was 14 um or 0.014 mm that can be entered in the pixel size of your camera calibration window...

The fiducials are important to correctly reproduce the internal geometry of the camera... but you would need the camera calibration report where fiducials coordinates would be reported ( something in range for x, y of +- 115 mm aprox. relative to photo center)

[cfc18]

I found a camera calibration report for this camera number. The report isn't for this project specifically, but it does have various x and y coordinates. I've attached it below. Would this work?

If so, would my next steps be to build an entirely new model with the images with fiduciary, mask the fiduciary is Agisoft, and input the coordinates?

[Paulo]

Yes, that is it....

you should follow the procedure indicated in the help menu on scanned photos....

However, it seems the coordinates of fiducials are relative to principal point of autocollimation... so i think you should subtract Indicated fiducial center coordinates corner coordinates from fiducials 1 to 4 coordinates and indicated fiducial center, midsize coordinates from fiducials 5 to 8 coordinates as in attachment....

This should help, but as pointed out by Matt, with good GCPs, you should not experience bowl effect....

[xabierr]

Hi @Paulo,

Just curious about using fiducials vs relying on self-calibrated camera parameters based on GCPs... is there a huge difference?