Understanding Image Residuals

[AndyC]

Please can anyone clarify exactly what the image residuals presented in the camera calibration section of the processing report indicate?

I believe it is (1) the amount of distortion (error) corrected by the lens model. However, I was also wondering whether it might instead be (2) the amount of distortion (error) that remains uncorrected after the lens model was applied?

(Apologies if this has already been answered elsewhere, I wasn't able to find this explained)

[piab]

Yes! Great question! Have been wondering this myself. Anyone have any ideas?

[Alexey Pasumansky]

Image Residuals figure shows the average vector of the reprojection error for the pixels in the corresponding cells. So averaging is across all the images in the calibration group and all the pixels inside the cell.
This image shouldn't be mixed with the distortion plot accessible from the Camera Calibration window after right-clicking on the calibration group label.

[AndyC]

Thanks Alexey,

I think I was a little confused because this is output was directly below the title 'Camera Calibration' and above the parameters of the lens calibration model in the processing report .
 
I can't seem to access the distortion plot you described from within the camera calibration window - Is this something new to the latest version (I've still using Version 1.1.5 build 2034)?

Thanks,
Andy

[Alexey Pasumansky]

Hello Andy,

Distortion plot feature was introduced in the version 1.2. This window shows the distortions and the image residuals (like in the report) on two tabs.

[piab]

Thanks for your help Alexey,

Would you be able to give a little more information on interpreting these outputs?
(I've read the user manual, but I'm still a little confused)

Am I right in thinking that the reprojection error is an indication of model quality, and higher reprojection error indicates parts of the model which are less well modelled?

For example, I get the output below for one of the three cameras I use in my set-up. As only one image is used from each camera for reconstruction, I use pre-calibration (determined from an array of 30 images - these have a much smaller and evenly distributed reprojection error). 

- Is the blank space due to lack of tie points these parts of the images?
- Is the alignment/orientation of the reprojection residuals due to insufficient constraint in the X axis?

[ugo]

Hallo,
i don't know what means have the Residuals Graph. How should I understand the red and green lines in the garph?
Can someone tell me what that means?