Confusion with the Camera Trajectory example

[Mechazo11]

Hello @AlexanderFabisch

I have two queries regarding the Camera Trajectory example found here https://dfki-ric.github.io/pytransform3d/_auto_examples/plots/plot_camera_trajectory.html#sphx-glr-auto-examples-plots-plot-camera-trajectory-py

1. Why are we performing ax = ptr.plot_trajectory(ax, P=P, s=0.1, n_frames=10) operation when the main plotting loop for i, c in zip(key_frames_indices, colors): is using cam2world_trajectory = ptr.transforms_from_pqs(P) which converted the pose in (x,y,z,qw,qx,qy,qz) into a 4x4 SE(4) matrix? I saved my odometry data in (x,y,z,qw,qx,qy,qz) format to match with TUM dataset (if I am not mistaken how they saved their poses)

2. My camera's coordinate frame is defined as Z, X, Y where +Z refers to translating forward, +X refers to translating right and +Y refers to translating vertically down. Is there a setting that we can change so that the generated 3D plot is oriented in this coordinate frame?

With best, Mechazo11

[AlexanderFabisch]

Why are we performing ax = ptr.plot_trajectory(ax, P=P, s=0.1, n_frames=10) operation when the main plotting loop for i, c in zip(key_frames_indices, colors): is using cam2world_trajectory = ptr.transforms_from_pqs(P) which converted the pose in (x,y,z,qw,qx,qy,qz) into a 4x4 SE(4) matrix? I saved my odometry data in (x,y,z,qw,qx,qy,qz) format to match with TUM dataset (if I am not mistaken how they saved their poses)

I don't see a problem here? Using a transformation matrix at the interface of plot_camera is more convenient since most of the time I saw extrinsic camera parameters represented by a transformation matrix. A more convenient format to store pose trajectories is the position + quaternion format. pytransform3d offers a function to convert between the two formats, so everything should be fine.

My camera's coordinate frame is defined as Z, X, Y where +Z refers to translating forward, +X refers to translating right and +Y refers to translating vertically down. Is there a setting that we can change so that the generated 3D plot is oriented in this coordinate frame?

This is exactly how the coordinate frame is defined here. See here.

[Mechazo11]

@AlexanderFabisch , Regarding the 2nd comment, can we pass an argument that would flip positions of Y and Z in https://dfki-ric.github.io/pytransform3d/_auto_examples/plots/plot_camera_3d.html#sphx-glr-auto-examples-plots-plot-camera-3d-py ?

[AlexanderFabisch]

I am not yet convinced that it is actually a useful contribution and not just a weird special case. Could you give me some links to documents that use this convention (and maybe describe it in detail)?

[Mechazo11]

@AlexanderFabisch I found this StackExchange post which shows what I am trying to achieve in terms of visualization.

https://stackoverflow.com/questions/8178467/how-to-plot-the-camera-and-image-positions-from-camera-calibration-data

[AlexanderFabisch]

But again, y is downward, x to the right. I don't see the difference?

[Mechazo11]

@AlexanderFabisch, I think the pytransform3d methods that perform the conversion from transformation matrices to position,quaternion are indeed orienting the axes correctly. But the plotting tools are putting Z axis on the vertical which is making any forward translation of the robot show up as a change in height of the graph. Hence, I was looking for a way to swap the axes so that forward translation of the robot would cause a change in length of the graph. Anyways, I am making a 2d graph for the time being to visualize the output for the time being. Thanks for replying to my queries.

[AlexanderFabisch]

But the plotting tools are putting Z axis on the vertical which is making any forward translation of the robot show up as a change in height of the graph. Hence, I was looking for a way to swap the axes so that forward translation of the robot would cause a change in length of the graph. Anyways, I am making a 2d graph for the time being to visualize the output for the time being. Thanks for replying to my queries.

I think you are confusing position and orientation. The position along the z component indeed translates the camera frame along the z-axis of the world frame. The z-axis in the orientation still points forward along one of the horizontal world axes.