Show Posts
1
Camera Calibration / Calibration Divergence
« on: January 20, 2021, 04:30:21 PM »
Hello, I’m trying to calibrate my camera with the software for latter use in a project regarding Close Range Photogrammetry.
I'm confronting a Nikon D700 with a prime lens, against a newer Panasonic GX80 (MFT Mirrorless) with the kit zoom lens (H-FS12032).
I'm having a small divergence on the calibration results with the mirrorless and kit lens.
I suspect the problem could be with the sensor moving or the lens mechanism changing in between ON/OFF.
Camera Settings:
- MF (Then I go to the digital slider to set the focus position to the lowest value (for near subjects))
- RAW
- Stabilizer OFF
- Aperture priority
- FN 13 for proper depth of field
- ISO 200
- Mechanical Shutter with 1 sec delay
- All fancy improvements (Intelligent Dynamics, Intelligent Resolution, Long Shutter Noise Reduction, Diffraction Compensation, Shadow Compensation) OFF
My workflow is as follow:
- I turn the camera on.
- Put it on a tripod.
- Shortest focal distance (12 mm) with the lens rig.
- Focus Position to the lowest (for nearby subjects)
- Select show Chessboard on the software and zoom out until the square size is 1 [mm]
- Shot with remote shutter (wifi) at least 100 photos changing the LCD position (inclination, tilt, depth) as much as possible in the range of depth (10 cm - 40 cm from the sensor mark)
- Convert the photos with Adobe Digital Negative Converter
- Upload the photos and calibrate the lens fitting all the parameters (f,cx,cy,k1,k2,k3,k4,p1,p2,p3,p4)
- Repeat 20 times
- Calculate the weighted mean and weighted std of the results from the calibrations sessions using as weight=(1/σ^2), assuming that the errors are equal to σ.
Problem:
Cx,Cy,B1,B2,P3 and P4 diverge (evaluated on the weighted STD)
Questions:
1) Square size of the chessboard is too small? How does this influence the results?
2) DNG vs TIFF I tried one set and the results changed a bit. Which format should I use?
3) The assumption that the errors are equal to σ is OK? Are those Std OK (attachment file)?
4) I tried fixing some parameters (f,cx,cy,b1,b2). Then I tried only the Image-Variant (f,cx,cy,b1,b2). Then I tried to combine them (fixing k1,k2,k3,k4 and Image-Variant f,cx,cy,b1,b2). Also I tried the rolling shutter compensation. I obtained the same results as before. I'm afraid that I'm doing something wrong here. Could you advice me on this?
I'm confronting a Nikon D700 with a prime lens, against a newer Panasonic GX80 (MFT Mirrorless) with the kit zoom lens (H-FS12032).
I'm having a small divergence on the calibration results with the mirrorless and kit lens.
I suspect the problem could be with the sensor moving or the lens mechanism changing in between ON/OFF.
Camera Settings:
- MF (Then I go to the digital slider to set the focus position to the lowest value (for near subjects))
- RAW
- Stabilizer OFF
- Aperture priority
- FN 13 for proper depth of field
- ISO 200
- Mechanical Shutter with 1 sec delay
- All fancy improvements (Intelligent Dynamics, Intelligent Resolution, Long Shutter Noise Reduction, Diffraction Compensation, Shadow Compensation) OFF
My workflow is as follow:
- I turn the camera on.
- Put it on a tripod.
- Shortest focal distance (12 mm) with the lens rig.
- Focus Position to the lowest (for nearby subjects)
- Select show Chessboard on the software and zoom out until the square size is 1 [mm]
- Shot with remote shutter (wifi) at least 100 photos changing the LCD position (inclination, tilt, depth) as much as possible in the range of depth (10 cm - 40 cm from the sensor mark)
- Convert the photos with Adobe Digital Negative Converter
- Upload the photos and calibrate the lens fitting all the parameters (f,cx,cy,k1,k2,k3,k4,p1,p2,p3,p4)
- Repeat 20 times
- Calculate the weighted mean and weighted std of the results from the calibrations sessions using as weight=(1/σ^2), assuming that the errors are equal to σ.
Problem:
Cx,Cy,B1,B2,P3 and P4 diverge (evaluated on the weighted STD)
Questions:
1) Square size of the chessboard is too small? How does this influence the results?
2) DNG vs TIFF I tried one set and the results changed a bit. Which format should I use?
3) The assumption that the errors are equal to σ is OK? Are those Std OK (attachment file)?
4) I tried fixing some parameters (f,cx,cy,b1,b2). Then I tried only the Image-Variant (f,cx,cy,b1,b2). Then I tried to combine them (fixing k1,k2,k3,k4 and Image-Variant f,cx,cy,b1,b2). Also I tried the rolling shutter compensation. I obtained the same results as before. I'm afraid that I'm doing something wrong here. Could you advice me on this?
