Obtaining the camera matrix

[egrigokhan]

I'm trying to obtain the camera matrices for each view in order to construct a visual hull for the objects but I have so far only been able to find the K matrices for each camera. I was wondering how I can get the R and t matrices for each view from the calibration data.

[feliyur]

Hi,

calibration['H_NP{ii}_from_NP5'] is (somewhat unintuitively, but that's the dataset's notations) the transformation matrix from NP{ii} to NP5 coordinate system. Likewise, poses[kk]['H_table_from_reference_camera'] is the transformation matrix from table coordinate system to NP5. Thus denoting the former T1 and the latter T2, the relative pose for view kk of camera ii is inv(T2)*T1, and from there you can extract R and t.

Hope this helps.
For myself I improved the wrapper functionality using gtsam, but didn't include it in the repo to avoid it as a dependency. If you're using gtsam I can send you the extended version.

Yuri

[egrigokhan]

Thanks a lot!