Explanation of Lens calibration?

[Rosy]

Hi,

I got some results of lens calibration (please see attached picture). But I have difficulties to interpret them.

Does any one know what the red or blue color represents, respectively? Are the red ones "bad" points?

What do the red or blue LINEs mean?

What about the white line? Is it an outlier?

How should I evaluate such an result. Should I remove the images with bad results and re-do the calibration?

Thank you guys!

[Alexey Pasumansky]

Hello Rosy,

Red and blue circles represent the detected corners of the cells. The color is demonstrating that everything is detected correctly: the circles cannot have "neighbors" of the same color. The lines represent the estimated position of the point according to the solved distortion coefficients.

[Rosy]

Hi,

Thank you very much for the explanation! I'm afraid I have a few more questions:

1. What does the direction and length of the lines (blue and red) mean? I see they have different length and directions.

2. For some result images, there are white points and lines. What does the white color represent?

3. Should I remove some images based on the results (like points/line color, length of the line, or the detected number of points) and re-do the calibration?  How to decide which images should be removed?

[Rosy]

The white arrows are shown in this attachment (marked with yellow circles).

What do they mean? What should I do then?

Thank you very much in advance!

[Alexey Pasumansky]

Hello Rosy,

White points are filtered out automatically and are not used for the lens distortion estimation.

[Rosy]

Hi Alexey,

Thank you very much for your answer. I only have one question left from previous post:

1. What does the direction and length of the lines (blue and red) mean? I see they have different length and directions.

BR,
Yuqian

[Alexey Pasumansky]

Hello Yuqian,

The lines represent the direction and magnitude of the distortions for the given point.

[Rosy]

Thank you very much!

[khouji]

The reason is simple. Manufacturing relies on producing a high quality product, not a perfect product. Could Canon and Nikon produce perfectly matched cameras and lenses? Yes, but none of us could afford them. So in order to keep cost down they manufacture equipment to fall within “Allowable Tolerances”. In other words if a lens back focuses just a little, or a camera front focuses a tiny bit that is ok, because the amount of time and money it would take to get things perfect is just not worth it.

[dysano]

Hi, I made a couple of lens calibration measurements and there are differences between the measurements (picture attachment). Are these differences significant? Thank you.

[Paulo]

Hi dysano,

you should look at Distorsion Plot (use right button context menu for selected camera group in Camera Calibration Window) and look at Correlation Tab. That way you can see how correlated your different camera parameters are for each calibration. see attachment....

Is this a checker board close range calibration or an in flight calibration?

[dysano]

Is is checker board close range calibration. Mobile phone Xiaomi mi8. The differences from the previous post will probably not be very significant. I dont understand Correlation tab. Can you explain it to me or some link where it is explained?

[Paulo]

Hello dysano,

further explanation on correlation off diagonal elements can be seen in https://agisoft.freshdesk.com/support/solutions/articles/31000158119-what-does-camera-calibration-results-mean-in-metashape-

High correlation  between parameters (gt than 0.5 max is 1 complete correlation) indicates that these parameters cannot be reliably determined. In previous post attachment, we see a correlation of 0.7 (high) between f and cy parameters indicating that may not be reliably determined... (small variance in f will highly affect cy).