Log in or register to post comments

How to recognize things like street signs?

May 4, 2017 - 12:52pm #1

I'd like to be able to recognize signage like stop signs out in the world. It seems like image recognition is too sensitive for this- it needs more detail than a sign has and it has to be the exact image. So while, under the right conditions it might recognize one particular picture of a stop sign, it would not recognize the actual sign or any other for that matter.  Text recognition works, but then you don't get the planar info. What would really be useful is a machine learning tool that can use a corpus of images like google image does.  Any idea how one might go about integrating the two or any other viable workaround? 

Thanks! 

How to recognize things like street signs?

May 6, 2017 - 1:25pm #3

Oh cool! Thanks! 

How to recognize signs in the wild?

May 5, 2017 - 2:13pm #2

Hello,

Actually, Vuforia's Text Recognition feature does return a 6 DoF pose. You can observe this in our sample app as the bounding box will change based upon the camera's orientation to the text.

However, as text is a sub-optimal target for our tracker, the pose will not be as accurate as it is for 2D planar targets.

Thanks.

Log in or register to post comments