Coordinates transformation from opencv camera frame to AirSim NED for object localization

[marcociara379]

Hi everyone,
I am trying to detect an Aruco marker in AirSim and localize it in NED frame after detecting it in the camera frame.
To check the correctness of my computations, I am comparing the estimated pose in quaternions with respect to the ground truth output by AirSim. Unfortunately, it seems that I have did something wrong and the two values mismatch.
Here is my procedure:

1. I compute the rvec, tvec rotation and translation vectors with cv2.aruco.estimatePoseSingleMarkers(corners, aruco_length, camera_K_mat, dist_coeffs). These should refer to the transformation from the Aruco frame to the OpenCV camera frame, that is the pose of the aruco marker in camera coordinates (cam frame should be x=east; y=down; z=north; aruco frame is x:east, y:north, z=out of the marker).
2. I convert the rvec rotation vector in rotation matrix: R_aruco_cam_, _ = cv2.Rodrigues(rvec)
3. I swap the axis of the R_aruco_cam_ matrix in order to get the aruco position in NED (from camera EDN): R_aurco_ned = np.array([R_aruco_cam_[:,2], R_aruco_cam_[:,0], R_aruco_cam_[:,1]])
4. I convert the rotation matrix R_aurco_ned to quaternions

The aruco marker pose from client.simGetObjectPose() does return a different orientation, unfortunately.
The quaternion conversion should be correct.

[ShrutheeshIR]

Hi @marcociara379
Were you able to solve this issue? I am facing a kind of similar issue. Does simply swapping the axes do, or I'm not sure if a negative sign comes somewhere. Let me know what you did to solve this issue
Thanks!

[nakole2]

Hi @marcociara379
Were you able to solve this issue? I am facing a kind of similar issue. Does simply swapping the axes do, or I'm not sure if a negative sign comes somewhere. Let me know what you did to solve this issue
Thanks!

Hi @ShrutheeshIR, did you find the solution for it? I am stuck on the same problem.