Hi,
Firstly, our image tracking (and most other AR systems that I’m aware of) only use a greyscale view for tracking, so switching colours doesn’t help a lot to distinguish images. That said, swapping dark/light colours would provide a visible difference to the tracker.
Image tracking tries quite hard to track, even when the camera image doesn’t match exactly - for example due to partial occlusion (user’s hand covering up part of it for example), shadows, imperfect printing, non-flat printout, etc.
That’s generally what you want when you know which target is being looked at, but isn’t helpful when you’re trying to work out which one of a set of images in view when the images have similarity. As an extreme example if 2 images are the same apart from one corner, then you want it to use that corner to “tell them apart” but to continue to track if that corner is covered up by the user’s hand.
Historically at Zappar we’ve used zapcodes to solve the “which target is this” problem, so our tracking is designed toward trying really hard to find a known target in an image rather than trying to solve the “are you sure it’s this target” problem. Zapcodes are not yet available through Universal AR though, so that’s not a solution you can use right now.
We are working on improvements to image tracking to make it better suited for telling similar targets apart, and multiple target applications in general (9 independent trackers isn’t great from a performance point-of-view, though is the only way to handle multiple targets for now).
Until then you unfortunately can’t rely on a target being reported as “seen” in a single frame as a reliable indicator of the target actually being there. You could monitor how many frames it is found for vs the others, wait until it is reported at a certain size in the image (close-up is less likely to mis-match), or encourage people to scan straight on and only accept targets where a vector pointing straight out of the target is within 15 degrees or so of the camera viewing direction.
Alternatively as you suggest you can try to add more differences between the images. Text often provides good strong unique texture, perhaps you could add a number to the targets too? The 3 images you’ve shown look like they should be “different enough” to me though.
Hope that helps a bit!
