Calibrated Camera vs PinholeCamera

[Peter Milani]

Hi All,

The documentation mentions that CalibratedCamera is preferred over applying a generic K calibration to a PinholeCamera. It says that it better to calibrate the measurements.

From my understanding of what this means is that Calibrated Camera uses a calibration matrix which is identity, but is there a class that can take a measurement from a generic K calibrated and reproject it into the identity calibration space?

cheers

Peter

[Dellaert, Frank]

Hi Peter

If you don’t know the calibration you minimize 

\sum |z-K.uncalibrate(pose.transform(point))​|^2 

If you already know the calibration, you can use

\sum |K.calibrate(z)-project(pose.transform(point))​|^2 

where K.calibrate needs to be done only once. Saves a ton of compute

Look for calibrate​ method of whatever calibration model you think of using.

FD

--
You received this message because you are subscribed to the Google Groups "gtsam users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gtsam-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/gtsam-users/291afb54-fe78-4fed-9cc9-75c7d899df9dn%40googlegroups.com.

[Peter Milani]

Thanks Frank, I've got it, but just confirming (and  dumbing it down for me): 

if z is the pixel coordinates (u, v), K.calibrate(z) transforms these into [x,y] coordinates which are x/z and y/z or x, y where z is 1 (through the application of the intrinsic matrix in the case of Cal3_S2).  Therefore, in the CalibratedCamera class I should pass x y as my measurement. Makes sense.

cheers

Peter