Compute reference maze corner points per default via closed corners?

[DennisDoll]

No description provided.

[KonKob]

Difference in pixel_length used as conversion factor based on detection of closed corners (c, n = 6) or open corners (o, n = 18)
Difference of means = 0.0033 px/cm
Difference of max and min conversion factor using closed corners is 0.021 px/cm

Conclusions:

conversion based on open corners seems more stable between recordings and should be preferentially used to calculate conversion
does conversion to cm differ too much between recordings?

[KonKob]

As one could expect, using the open corners is less variable than using the closed ones. (not in our case, but given, that the DLC tracking is stable)
Obviously, if you have a small error on 4 cm and multiply it by 12.5, the error increases by 12.5. On the other hand, if you have 50 cm for normalisation, the error doesn't increase any more.

So, whenever the tracking of the open corners is stable, I would only use the closed ones as last possibility.