Log in or register to post comments

Detecting 3d objects

June 12, 2012 - 2:32am #1

Hello,

I'm interested in recognizing 3d objects having irregular shape, for example two different shoes.
I'm only interested in recognition and not tracking, so the pose is irrelevant in my case.
I know that QCAR doesn't support it directly, however I had an idea and I'd like to consult with you to see if such thing might somehow work before I start working on a proof of concept.
If I take pictures of the shoes from many different angles (for example 20 for each one) and create in total 40 trackables, do you thing it might recognize the object from all (or at least a wide range of) angles?
What I'm not sure of is whether images that are too similar might cause a problem for creating a dataset of trackables.

So - what are my chances?

Thanks.

Re: Detecting 3d objects

June 12, 2012 - 3:46am #2

Hi maxh

I think this topic has come up before over here:

http://ar.qualcomm.at/node/2000418

I think Kim summarised it very well:

"You cannot detect arbitrary 3D objects, but you can detect 3D objects made up of planar image targets (e.g. a cereal box). Look for the MultiImageTargets section of the Developer Guide and the AR Extension for Unity 3 documentation (https://ar.qualcomm.com/qdevnet/sdk/unity/ar). You can create simple cube objects using the My Trackables system, or you can edit the config.xml file by hand to arrange image targets into the desired configuration."

So basically you will most likely find it very difficult to get your proposed approach to work, and if it did then reliability would be an issue due to factors such as similarity, shadows etc.

HTH

N

Log in or register to post comments