I´m wondering if it is possible to make a marker out of a label that is stuck on a round bottle.
While surface reflection and refraction will always be concerns in feature-based image recognition, this is an example where the feature distribution within your bottle's label would need to be very uniformly distributed to prevent feature occlusion from breaking the experience entirely. Most times a bottle will have a linear (or as close to linear as possible) hotspot that forms due to real-world environment lighting and the user's viewing angle in relation to the bottle. If your bottle's label is relying on a heavy concentration of features that, if occluded would break the experience, then you may want to add more unique features and be sure the feature distribution is uniform to compensate for inevitable occlusion due to user interaction, shadows, lighting, etc.
However, I understand that you may not have control/influence over the bottle's label due to branding restrictions, clients, etc. In this case, strive to get them the best experience possible while working within the limitations that all feature-based image recognition AR experiences are plagued by: lighting, occlusion, and user error!
I agree with you vincekilian, thanks for pointing it out.
In fact, we've had better results with single image than with multi-targets, being that the multi-target result in several narrower images that are on their own harder to track. This said, once acquired: multi-target have WAY better tracking response to the 3d object being moved around than the single 2d image, there's a lot less jitter or loss of tracking.
Something else that is bothersome: reflexion. On a curved object such as a bottle, there is often a glare line somewhere, that reduce recognition. It would be great to be able to reduce it by filtering, or play with exposure... but I've already mentionned it in the feature request thread.
Single targets can be used if your source image used to create the trackable image accounts for the ocular distortion caused by the bottle's curvilinear surface.
It means your users generally needs to trigger the experience from as close to "straight on" as they can, but after the initial image recognition (OnTrackingFound()), you can instantiate a prefab with geometry modeled to replicate the curvilinear surface normals of the bottle.
The multi-target method could allow for less accuracy in the initial pose required of the user to "trigger" the experience, but I've acheived this with transparent/alpha video playback as well as standard video playback using a single standard image target with surface distortion factored in.
You're welcome. If you revise the XML configuration file in the project, be sure to select Vuforia > Apply Dataset Properties from the menu afterwards. This will update the MT instance in the scene to reflect your changes.
Thanks a lot for the hints. I'll experiment with this for sure.
Even an approximation should lead to better results than having a straight flat target over a rounded surface; sometimes that's all that we need :D
Yes you can omit MT faces, you can also revise the XML configuration file to customize the geometry of your child target arrangement.
The technique that you describe, using multiple faces to construct a curved surface is the same method that we advise, though it takes some testing and experimentation. The MT target type still requires a flat child target, and so any curve fitting is an approximation at best.
I tought multitargets were only for square/rectangular boxes.
Can we omit top/bottom faces? Can we alter the square-box base of the multitarget to obtain something more suitable with a round bottle? Maybe 6-8 faces for some hexagonal- or octongonal-based prism or something?
If we have only one side that counts (in this case a bottle label) can we omit "the rest" (having say 3 partial faces oriented together)
I haven't experimented much with multitargets yet but am currently looking ar rounded-surface trackers, this could come in handy.
It's not recommended, though some parties have done this. The trick is to divide the target face into sections and define it as a multitarget so that the curvature of each face is minimized.
Are you sure you want to delete this message?
Are you sure you want to delete this conversation?