Add Orientation #83
Add Orientation #83
Conversation
Adimote
force-pushed
the
add_new_vision_features
branch
from
7f39f10 to
a4d2705
Compare
Adimote
force-pushed
the
add_new_vision_features
branch
from
a4d2705 to
0e6e164
Compare
robot/markers.py
Outdated
| Rotations around the different axes can be thought of as follows: | ||
| - X rotation represents pitch. Positive rotation | ||
| around the X axis represents a rotation with the nearest point going downwards, | ||
| and furthest going upwards. (i.e. a bowing motion is a positive x rotation) |
There was a problem hiding this comment.
It took me a couple of attempts to understand the second sentence here ("Positive rotation..."). I like that that approach applies quite easily to each of the descriptions, though it's a bit complex on first reading.
I wonder whether moving the example first would help people grok the idea sooner so that the technically-correct explanation then has an idea to hold onto?
Alternatively, is there a slightly different phrasing which is easier to grok first time?
(I'm assuming we'll use a similar phrasing in the main docs too, so I think this is worth getting right here)
robot/markers.py
Outdated
| right. | ||
| - Z rotation represents roll. Positive rotation around the Z axis represents | ||
| a rotation with the highest point going to the right, and the bottom going to the | ||
| left. |
There was a problem hiding this comment.
Z rotation is aligned with a clockwise rotation about the axis (when viewed from the positive end of the axis), yet the other rotations are counter-clockwise. Is the difference intentional?
I'm not sure whether this matters, just noting it as my assumption was that they'd be consistent.
There was a problem hiding this comment.
The Z axis actually comes towards you, so its still a counter-clockwise rotation around the axis.
There was a problem hiding this comment.
The Z axis actually comes towards you,
You're right that I was wrong, though I'm not sure I agree with this statement -- positive Z is away from the camera.
so its still a counter-clockwise rotation around the axis.
Indeed.
There was a problem hiding this comment.
Further to my being wrong here -- I think that it's Y which is actually clockwise; might this be a result of our y-coordinate being defined with a different positive direction to that of OpenCV?
|
|
||
| class Orientation(_Orientation): | ||
| """ | ||
| Represents the orientation in 3d space as rotations around x, y, and z axes. |
There was a problem hiding this comment.
What is the order of application of the angles, and/or do they always apply to the cartesian axes of the camera?
A camera-relative interpretation of (45°, 90°, 0) is as follows: consider a marker which is leaning forwards by 45° but placed on a horizontal rotating platform (like a "lazy susan"); then rotate the platform by 90°.
A marker-relative interpretation is as follows: consider a marker perpendicular to a rotating platform which is angled 45° towards the observer (like a lazy susan on the surface of a table which has been half-pushed over); then rotate the platform by 90°.
In both scenarios the marker is now facing off to the side, however in the former its baseline is still horizontal while in the latter its baseline is at a 45° to the horizontal.
I'm not sure which of the two systems is easier to think about, nor which we've got data for. We should find out and document this however.
Note that if we apply the angles in the reverse order (z then y then x) then the two scenarios above change roles.
There was a problem hiding this comment.
@Adimote I see this is partially resolved by 6018425, thanks for that, though we still don't describe whether the rotations are relative to the marker or the camera.
I think I hope that they're relative to the fixed frame of the camera (I'm expecting that that would make the angles for a pair of markers more easily comparable), though I've not explore enough examples to know which is better/easier to think about.
There was a problem hiding this comment.
Took me a while to figure this out, but I'm pretty sure the rotations are actually applied relative to the marker. I've added this detail in 7535af1 This means an upside-down marker would report inverse rot_y. We might want to investigate how to make this easier for the students to use in a later update.
robot/markers.py
Outdated
| Describes a rotation as angles about the x, y, and z axes, which | ||
| corresponds pitch, yaw, and roll of the marker respectively. The | ||
| angles are measured as an offset from the marker directly facing | ||
| the camera. |
There was a problem hiding this comment.
I think the last sentence here should have "a marker" rather than "the marker".
|
Marked this as WIP because it's not ready for shipping due to the confusing output values |
Add Orientations to markers, to expose the feature added in sourcebots/sb-vision#34