From camera frame to world frame using habitat.utils.geometry_utils

[yhymason]

Hi there, I am trying to convert coordinates of point cloud coordinates which are initially in the depth camera's coordinate frame back to the simulator's world frame. I am not sure what's the correct way to do it here since I am not familiar with the use of quaternions in Habitat. I have the following 2 transformation logic:

            points_world[idx] = (
                quaternion_rotate_vector(depth_camera_rot, points_world[idx])
                + depth_camera_pos
            )

or

            points_world[idx] += depth_camera_pos
            points_world[idx] = quaternion_rotate_vector(
                depth_camera_rot, points_world[idx]
            )

where depth_camera_rot is the quaternion of the depth camera, and depth_camera_pos is the world coordinate of the camera.

        depth_camera_pos = np.asarray(depth_camera_info.position)
        depth_camera_rot = quaternion.as_quat_array(
            np.asarray(depth_camera_info.rotation)
        )

Can you help me confirm which logic will work?

[erikwijmans]

Rotation + translation works the same with quaternions and matrices. First apply the rotation then apply the translation.

To create the quaternion from a list, please use habitat.utils.geometry_utils.quaternion_from_coeff as there are different ways to store the coefficients of a quaternion and that'll make sure it's consistent.

[rpartsey]

@yhymason Feel free to re-open the issue, if you still have questions.

[dhruvbatra]

@rpartsey — we get this question about Habitat coordinate system quite frequently. We should create a page or tutorial on it with visuals so that we can point people to it. I believe Eric has a draft PR on it that you might want to take over.

[rpartsey]

@dhruvbatra Thanks! That's exactly what I planned to do. Opend a new Issue https://github.com/facebookresearch/habitat-lab/issues/954 and assigned to myself.