[calibration-graph-cli] Pinhole calibration bug

When calibrating a camera with a pinhole model, the calibration-graph-cli estimates the distortion with 5 parameters
“distortion”: [-0.3046111298320273,0.19613377571040838,0,0,-0.14857597315214646],

When exporting the graph to rig, only the first three are exported to the rig:
“distortion”: "-3.04611116647720e-1 1.96133777499199e-1 0.000000000000000 ",

This seems like a bug, especially because the value of the last parameter is significant (-0.1485759731521464)

Questions:

  1. Which of the above coefficients correspond to the radial and which to tangential distortion?
  2. In dwPinholeCameraConfig there is support for a single distortion polynomial (I presume radial distortion)?
  3. If the calibration estimates both radial and tangential distortion, it seems wrong to use just radial distortion when rectifying the images.

I am attaching the related files, please check the CALIBRATION_LOG.log for the calibration log with appended calibrated graph and a rig file:
CALIBRATION_LOG.log (844.8 KB)

Thanks in advance!

Hi vzadrijaRA,

Just wanted to let you know that we are looking into this and discussing with the calibration team. Will update the thread when we get more info.

Hi vzadrijaRA,

Thanks for helping us identify this bug. We have reported this to developers and expect it will be updated in a future SW release. In general we recommend that you not use the pinhole model and instead use the ftheta model. Our experience is that it is better at modeling geometric properties of the sensors we are supporting.

To answer your questions:

1. Which of the above coefficients correspond to the radial and which to tangential distortion?
We are not handling tangential distortions at the moment.

2. In dwPinholeCameraConfig there is support for a single distortion polynomial (I presume radial distortion)?
Yes, only radial distortion polynomial of length 3 for pinhole