[WIP] fix for rotation X bug

[kalwalt]

⚠️ All PRs have to be done versus 'dev' branch, so be aware of that, or we'll close your issue ⚠️

What kind of change does this PR introduce?

Fix for issue #59 missed rotation on X axis.
Can it be referenced to an Issue? If so what is the issue # ?
issue #59
How can we test it?

Test the basic and dev exemples (three.js)
Summary

The object 3D is not rotated correctly, we should apply a rotation on the X axis to fix the problem.
Does this PR introduce a breaking change?

Please TEST your PR before proposing it. Specify here what device you have used for tests, version of OS and version of Browser

Other information
A said in the issue #59 this bug was introduced while deleting the tango and aruco code, we (me) deleted a relevant part of the code

https://github.com/jeromeetienne/AR.js/blob/024318c67121bd57045186b83b42f10c6560a34a/three.js/src/threex/threex-armarkercontrols.js#L130-L135

Applying again the rotation X will solve the issue.

List of features / to do list:

• fix for issue bug with rotation  #59 reapplying the rotation matrix X to threex-armarkercontrols.js
• upgrading artoolkit version for ar.js and aframe-ar.js ( previuosly only for ar-nft.js, aframe-nft-ar.js) This will be done in another PR.
• removing the multiplication for the modelviewMatrix with ProjectionAxisTransformMatrix see line 258 in threex-artoolkitcontext.js
• fix examples (three.js and aframe) for the position of the 3d objects (y instead of z)
• removing old artoolkit.min.js libs This will be done in another PR.
• removing references/links to the old Ar.js repository
• testing the examples on Desktop and Mobile device.

[kalwalt]

Empirical demonstration with dev.html example, threex.artoolkitcontext.js without
projectionMatrix.multiply(this._artoolkitProjectionAxisTransformMatrix)
and artoolkit-nft.min.js

dev.html example threex.artoolkitcontext.js with
projectionMatrix.multiply(this._artoolkitProjectionAxisTransformMatrix)
and artoolkit.min.js

Note the three axes helper is oriented in the same way in both the images.

[kalwalt]

I did also this gist for proving the theory https://gist.github.com/kalwalt/528e722706847ea523c489947d07bc8c

[kalwalt]

arjs-session.html example without multiply X matrix:

with the right multiplication matrix, working as should:

the white grid must be in the same plan of the marker.
In the next hours will check also the nft examples (threejs and aframe).

[kalwalt]

I added the rotation X matrix also to the arMarkerControls-nft, because you could use in combination with a Pattern Marker. Soon i will provide a codepen example.

[kalwalt]

It's possible but i don't reccomend to work with both, test this codepen: https://codepen.io/kalwalt/pen/LYVrQKd test with an Hiro marker and the trex image. When pointing to the hiro marker you will see two blue cube flashing on the lefte and after on the right. I think we should find a solution also for this, not of course in this PR.

[kalwalt]

Testing the threejs examples with Mobile, the cube is distorted in portrait mode:
instead in landscape mode works as expected.
This happens for almost the examples. default-thinner-border.html and test-runner-html not because use an image as source.

[kalwalt]

We get different Matrix from Camera because this changes in Artoolkit5 arglCameraFrustum was removed in favour to arglCameraFrustumRH in this commit see how i changed the code in ARToolkitJS.cpp
artoolkitx/jsartoolkit5@b3b9682#diff-e307757acae7e50b4037668bc9d31b32

[kalwalt]

I'm testing if adding back the old arglCameraFrustum method has some effect on the code
my testing branch https://github.com/kalwalt/jsartoolkit5/tree/fix-arglCameraFrustum .
This not change anything on Mobile Device. You get the same result, and most important. you will have the same result if you use the old artoolkit.min.js without artoolkit.api.js. So artoolkit.api.js is necessary because add some extra code (and probably was modified for this reason?) to solve the ratio issue.

[kalwalt]

For sure is relative to changes in artoolkit.api.js see this commit in the old AR.js repo, there are changes in _copyImageToHeap to the portrait mode. Maybe we should make this changes in Ar.js?

[kalwalt]

I found that if you print the orientation from arController give wrong result:

arToolkitContext.init(function onCompleted(){
		console.log(arToolkitContext.arController.orientation);
// other code here
}

in portrait mode should give portrait instead give landscape
This not happens in jsartoolkit5 it gives me the right result. I think i will open another issue for this.
Because it works well under jsartoolkit5 we have found the reason of this.

[kalwalt]

I think that is better to merge the fix for the bug in the Rotation on the X axis, and after make the upgrade to the new artoolkit lib. This PR need testing anyone want to test is welcome! 🙂

[kalwalt]

Hi @fcor have you tested this PR?

[kalwalt]

For anyone wants to test this PR try this codepen: https://codepen.io/kalwalt/pen/ZEbYwdq you should see a red cone onto an Hiro marker: the corner pointing outside the marker and placed correctly on it.

[nicolocarpignoli]

@kalwalt please write on CHANGELOG file, directly on dev branch, this modify :) thanks!

next version will be 3.1 so write under 3.1

[kalwalt]

@kalwalt please write on CHANGELOG file, directly on dev branch, this modify :) thanks!

next version will be 3.1 so write under 3.1

@nicolocarpignoli i will update the CHANGELOG 🙂