Intrinsic Calibration

Intrinsic calibration

TagSLAM can not perform intrinsic calibration. You must supply the intrinsic parameters in the cameras.yaml file.

If you are dealing with a “normal” lens, it is recommended that you use the ROS camera_calibration package for intrinsic calibration. It can do the standard radtan (plumb_bob) radial-tangential distortion model and is the easiest tool to use. However, it’s output file must be converted to work with TagSLAM

If you are dealing with a fisheye (equidistant) lens, consider using Kalibr or the multicam_calibration tool. Kalibr and multicam_calibration will already give you the output in a format that is close to the one that TagSLAM uses.

Converting ROS calibration files

After running the ROS calibration you will end up with a file that is called ost.yaml. It will look like this:

image_width: 1920
image_height: 1200
camera_name: pg_17408093
camera_matrix:
  rows: 3
  cols: 3
  data: [2168.758281, 0.000000, 990.688651, 0.000000, 2166.979687,
583.914090    , 0.000000, 0.000000, 1.000000]
distortion_model: plumb_bob
distortion_coefficients:
  rows: 1
  cols: 5
  data: [-0.189067, 0.124915, 0.000043, 0.000296, 0.000000]
rectification_matrix:
  rows: 3
  cols: 3
  data: [1.000000, 0.000000, 0.000000, 0.000000, 1.000000,
0.000000, 0.000000    , 0.000000, 1.000000]
projection_matrix:
  rows: 3
  cols: 4
  data: [2094.321777, 0.000000, 993.054257, 0.000000, 0.000000, 2136.192139, 583.488844, 0.000000, 0.000000, 0.000000, 1.000000, 0.000000]

This file must now be converted to a format that TagSLAM can process:

rosrun tagslam camerainfo_to_kalibr.py -i ost.yaml -o cameras.yaml

The resulting file should look like this:

cam0:
  camera_model: pinhole
  intrinsics: [2168.75828, 2166.97969, 990.68865, 583.91409]
  distortion_model: plumb_bob
  distortion_coeffs: [-0.18907, 0.12491, 0.00004, 0.00030, 0.00000]
  resolution: [1920, 1200]
  tagtopic: /pg_17408093/tags
  rig_body: XXX_NAME_OF_RIG_BODY

You will need to adjust the tagtopic to match the actual ROS topic under which the extracted tags are published. The rig_body must be set to match the name of the TagSLAM body to which the camera is attached.

NOTE: the intrinsics data is [fx, fy, cx, cy]! It is a common mistake to switch that around, as they show up in a different order in the K-matrix.