29 September 2014

I’ve tried working with ArUco in ROS, but for some reason I don’t get a correct position back. Good thing is, I found Sahloul’s ar_sys!

Sometimes a marker’s Tvec is approximately [0, 0, 0], sometimes to enormous values (~ ±1e15, z seems fixed to 9007199254740992.0).

Things that might be a problem:

• Extrinsic values are not computed for the markers at all.
  • Test: Read source code for aruco::Marker and aruco::MarkerDetector.
  • Result: To compute the extrinsic pose of the marker, aruco::Marker::calculateExtrinsics should be called. This is done in aruco::MarkerDetector::detect iff the camera matrix (either passed directly, or through a aruco::CameraParameters object) is set. This is done correctly.
• Camera does not know its parameters
  • Tests: Inspect camera’s parameters from the ROS topic /c920/camera_info
  • Result: parameters are the same as in the file ~/.ros/camera_info/c920.yaml
• ArUco does not know the (right) camera parameters.
  • Test: Print out the ArucoDetector::camera_parameters_’s fields.
  • Result: ArUco got the wrong values. I read in a cv::Mat of double values with CV_32F. However, those are floats. For doubles, use CV_64F. Still wrong position, while both parameter sets are the same.
• aruco::Marker::Tvec is not set correctly.
  • Test: Print out Tvec.
  • Result: It’s similar to the values that are output by ROS (minus the +/- transforms of aruco::Marker::OgreGetPoseParameters). No errors in converting from ArUco to to ROS.

Error resides somewhere in ArUco/parameter part, but not sure where! I will need help of Toby next meeting.

Update: I found Sahloul’s ar_sys by coincidence (googling for ROS AR packages). This package does work. A screenshot from Rviz: