Difference in yaw/roll/pitch correction

[PanierAvide]

Describe the bug

I have this image which causes a difference in rendering compared to Panellum : https://panoramax.openstreetmap.fr/images/bf/3f/be/8c/3a62-47f1-b744-6a75f9fff922.jpg

• PoseHeadingDegrees = 0.0
• PosePitchDegrees = -24.49
• PoseRollDegrees = 8.96

Panellum rendering

PSV rendering

Which is weird is that if I set manually sphereCorrection with negative values for tilt/roll, display becomes similar to Panellum 🤔

Also, this might be kind of specific, as these other image doesn't project differently with Panellum :

• Theta S with y/p/r : https://panoramax.openstreetmap.fr/images/db/20/81/99/10d9-4f32-869c-f5249194dac8.jpg
• Theta S with p/r : https://panoramax.openstreetmap.fr/images/23/4c/eb/be/2d4f-4294-8e13-2def46155b67.jpg

Although trying to understand deeper what could cause this difference by digging in the code, I can't find any obvious thing.

Online demo URL

https://codepen.io/PanierAvide/pen/zxvbjXb

Photo Sphere Viewer version

5.14.0

Plugins loaded

No response

OS & browser

Linux + Firefox latest

Additional context

Might be a follow-up of #1406 ?

Best regards

[mistic100]

As you already did a rather deep analysis, did you find in panellum source code where the pose is applied ? Just in case so I don't go read their code again :-)

[PanierAvide]

If I'm not mistaken, this should be here : https://github.com/mpetroff/pannellum/blob/7769cc66cb72d1a02c9aa1e818225c657d56de26/src/js/libpannellum.js#L924

[mistic100]

Funny. In the first iteration for #1406 I did added negative signs on roll and pitch, but it wasn't satisfactory either.

I feel there no real standard to the angles order and direction... If only they all used a Quaternion instead of Euler angles !

I'll make some investigation, perhaps there is pattern, and also check the whole pannellum implementation.

[mistic100]

I honnestly don't know what to do

• inverting the tilt/pitch angle only fixes your first image, but inverting also the roll give the opposite effect has pannellum
• your two other examples have very low pitch, and I think this is why you don't see a difference

and the values of these two last examples are completly wrong anyway (no permutation of minus signs give a satisfactory result)

[mistic100]

Perhaps just inverting the tilt is the right solution

It seems okay with this other example #1530

[PanierAvide]

Not sure if that helps, some control images with higher pitch/roll values that renders similarly in current PSV and Panellum :

• Heading/Pitch/Roll = 270/-20/-21 : https://panoramax.openstreetmap.fr/images/73/5d/32/3c/e557-41c4-941e-ce921f86108b.jpg
• HPR 0/-33/-22 : https://panoramax.openstreetmap.fr/images/a4/15/9a/02/8f9d-4f14-924b-9f223a257305.jpg
• HPR 22/42/-23 : https://panoramax.openstreetmap.fr/images/10/38/0c/5e/cb94-452c-8545-6c65bbe06c3c.jpg

[mistic100]

I have a problem with your first image: why does it a have non zero pitch and roll ? the image is already perfectly horizontal, applying any correction make it totally crooked.

[PanierAvide]

The forever fulfilling joy of crowdsourcing ? 🤣 In fact we have people testing some "centrales à inertie", it's probably an invalid value associated to the picture. In some future, we would allow users to correct interactively the pitch/roll to improve quality of these metadata.

[mistic100]

In order to make the appropriate changes I need a fiable source of data. As it is not really the case of files from PanoramaX, I made some simulations with Photoshop plugin Flexify 2. If you, or other people, can tell me what you think of each case.

@qhess34 @Fluidisator I think your input would be valuable as you worked with some IMU attachements :-)

---

Preamble: I cannot find a single fiable source specifying wich direction is positive and wich is negative. All diagrams have these stupid double arrows without any sign.

The most precise I get is https://en.wikipedia.org/wiki/Aircraft_principal_axes but does these apply to a camera ?

The good news is I a pretty sure the angles order is correct (ThreeJS YXZ = pan then tilt then roll), only some signs are incorrect.

---

This is the base image, totally flat

The next images have a tilt, roll or pan, I have have described them by the movement of the camera to achieve such transformation (so when the camera goes "up", the center of the panorama goes "down").

---

These are tilted up (A) and down (B), which one has a positive tilt angle ?

A:

B:

---

These are rolled counter-clockwise (A) and clockwise (B), which one has a positive roll angle ?

A:

B:

---

And finally these ones are panned to the right (A) and to left (B), which one has a positive roll angle ?

A:

B:

---

Another interesting question: considering a panorama with all angles at 0°, is the north in the center of the image or on the sides ? (I always assumed the north was at the center, and that's how the Virtual Tour, Map and Plan plugins work).

[qhess34]

Hello,

In my opinion, we can use this picture as reference for the axes (camera monitor is at the back)
https://github.com/qhess34/pan360-helmetfix/raw/main/doc/gopromax.jpg

⚠️ Note: The XYZ axes here (aeronautic system) are not the same as Three.js (3D graphics) axes. ⚠️

x=roll => ThreeJS z
y=pitch/tilt => ThreeJS x
z=yaw/pan => ThreeJS y

Concerning the signs, I suggest considering positive angles as anticlockwise.
So it give :

• Roll : Left ear on left shoulder : x -> -90
• Roll : Right ear on right shoulder : x -> 90
• Pitch : Heading to the ground : y -> -90
• Pitch: Heading to the sky : y -> 90
• Yaw : Tilting to the left : z -> 90
• Yaw : Tilting to the right : z -> -90

So, for your pictures:

Pitch (y):

• A: positive (camera looking at the sky)
• B: negative (camera looking at the ground)

Roll (x):

• A: negative (camera’s left face close to left shoulder)
• B: positive (camera’s right face close to right shoulder)

Yaw (z):

• 0° in the photosphere context corresponds to the direction of the camera that we want to display by default (the center).

[mistic100]

Thank you.

However I have a interpretation issue between these two sentences.

I suggest considering positive angles as anticlockwise

Roll : Left ear on left shoulder : x -> -90

For me, having left ear on left shoulder is rotating the head/camera anticlockwise, so is it positive or negative ?

(on peu échanger en Français si ça évite la barrière de l'Anglais :-) )

[mistic100]

I think my confusion comes from the fact that ThreeJS axes, beside the different names, dont have the same direction either.

This is using the AxesHelper at the origin and the fisheye feature to move the camera away. At init the camera is aligned with the blue axis. The X axis is red. The Y axis is green. The Z axis is blue.

And to make things worse, my illustration here is wrong (at least for tilt) https://photo-sphere-viewer.js.org/guide/config.html#spherecorrection