This is my own version of the multi markers.

Definition of multi marker.

a multi marker is a group of marker acting as one.

Goal

Make a multi marker with a nice workflow. It must be as easy as possible for the user. It must handle dynamic markers positions.

The user put all the markers where he want until he is satisfied. This is the dynamic part.
After that, markers positions is considered stable. aka they are no more supposed to move.
Those markers are called sub markers.
AR.js will build a multi marker based on them.
All the sub-markers will act as one, the multi-marker.
the multi-marker pose will be averaged from all the visible sub-markers.

Thus even if only one sub-marker is visible, multi-marker position will still be tracked properly

Area Learning Process

the user moves its phone around the area to learn
we detect markers position
we store statistics on the respective position between each marker
- if i simultaneously see marker1 and marker2, i store the position of marker2 in relation to marker1
When the learning process is considered done, output a file containing a description of this multi-marker

Motivation

tl;dr; there was too many bugs in artoolkit
issue in the file parsing - TODO get github issue
in jsartoolkit5, it has several issue in the multimarker file loading. It is impossible to use twice the same marker filename.
it handles poorly the url. It is forcing the user to put all the files (multimarker file plus all the pattern file) at the root of the url.I tried to workaround those issues, but it rapidely became a pain. So i prefered to devote my energy at building a good version instead of working around those bugs.

High Level Description

The workflow is splitted into 3 steps :

collect statistics about the relative position of each marker couple
generation of multi-marker description
using the multi-marker

So we first have a learning phase to collect data and generating result, then we have a usage phase, where the multi-marker is used in a AR application.

So in short, you put all your markers where you like them. Then you scan them with your phone to collect statistics about their relative position. When you are done, you start using the AR application with this multi-marker.

Definitions

a multi-marker is several markers acting as one
a markerA is said to be a sub-marker of a multi-marker

Learning Algo description

every time i see markerA and markerB at the same time, i store statistics on their relative position
such couple is called a seen-couple
the statistics collected are simply an average of position/quaternion/scale
this statistics can be trivially expressed as a transformation matrix

Generation of Multi-Marker description

when the learning is considered completed, we generate a multi marker description
- it is a json file in which each sub-markers has a transformation matrix relative to the origin sub-marker
the first sub-marker is the anchor/root sub-marker.
The origin position of the multi-marker is the origin sub-marker
the origin sub-marker MUST be seen at least once during the learning phase

Description Format

the description format is simple:
it is a list with one item for each sub-markers
- it contains a description of the sub-markers (e.g. marker barcode 3)
- and a poseMatrix which is the transformation matrix relative to the origin sub-markers
- the first item is the origin sub-marker

Computation of the average matrix for each sub-marker

Once statistics are collected, we need to produce a description of this new area. This is a json structure describing each sub-marker and their respective position.

Here are the algo which can be used

Minimal Algo - assume origin always visible

Assume that the origin sub-marker is ALWAYS visible
Look on all the seen-couple of the origin sub-marker
for each sub-markers, you got the relative position with the origin sub-marker
just generate the transformation matrix for it

This is the one implemented

Minimal Algo - origin may sometime be invisible

do first pass where you compute all couple which contain the origin submarkers
so you computed the transformation matrix for each sub-marker which has been visible with the origin marker
you computed the level 1 of transformation matrices
the level 0 would the transformation matrix of the origin sub-marker. But it is always the identity matrix.
you do an iterative process, where each iterations compute one more level.

Algo for one iteration

for each sub-markers, check all the seen couple it appears in
if at least one already has a transformation matrix, compute the transformation
Loop until all sub-markers got computed