GTSAM Camera model

[ujas mandavia]

Hello.

I have a pinhole camera lens with a 5 parameter radial-tangential distortion model k1, k2, k3, p1, p2. For this camera model I'm not getting the differnce between the usage of gtsam::Cal3_S2, gtsam::Cal3DS2 and gtsam::Cal3DS2_Base calibration models. WHich one is appropriate to use.

If I use gtsam::Cal3_S2 calibration model and it doesn't uses those radial-tangential distortion parameters, will my triangulation will be right with this calibration model without using those distortion parameters?

Thank you,

Ujas

[voltron]

An easier way to do this is to undistort the imagery before you pass it into your cv pipeline.

OpenCV has an undistort function. https://docs.opencv.org/3.4/da/d54/group__imgproc__transform.html#ga69f2545a8b62a6b0fc2ee060dc30559d

If you undistort the imagery first, you don't have to think about distortion. Then you use Cal3_S2.

If you are unable to undistort first:

Cal3DS2_Base provides calibrate/uncalibrate functionality.

Cal3DS2 provides the retract and localCoordinates functions in addition to calibrate/uncalibrate.

It looks like you use Cal3DS2_Base if you are not estimating the calibration and use Cal3DS2 if you are optimizing the calibration.