Start a new topic

Advice on object recognition

Advice on object recognition

Hi, 

I was wondering if there was a way of modifying the image recognition component of this SDK in order to track and overlay on an object. Bascially using a photo of item A to track and overlay over item A even with the varied lighting conditions, etc. Or will only the exact same image as that kept within the .wtc be tracked and overlayed upon.

Regards

the component is designed to track 2D images. If the object could be represented well by a flat image it is worth a try to see if it recognizes and tracks well.

we have tested e.g.  flat t-shirt logos, flat buidlings face

it highly depends on the quality of the target to track and how much changes are there in lighting. If you can give us more information on the type of object you want to track we might be able to give you addtional information and guidelines. However keep in mind that our solution currently works best with flat 2D like image targets.

i am working on recognition of small statues, i saved the images with white background in wc file. while recognizing when i place statues at some other places with diff background, app does not recognise. how can i solve this problem? 
Hi Aamir,

generally our image recognition solution is targetted towards detecting and tracking planar images. If the target object exhibits only a slightly three dimensional structure it may also work (e.g. an image on a t-shirt).

In order to better understand your use case and to give you assistance I would like to kindly ask you to post an example of
1) a target image you are trying to track
2) a sample of an image that tracks well (e.g. white background)
3) a sample of an image that does not track (e.g. differenct background)

Thank you very much for your cooperation.

Best regards,
Nicolas

Thanks Nicolas for your reply..

I am attaching two images, one image named as image.jpg is used in wtc file. When same image is tracked it gives result but when other image named as result.jpg is tracked it does not give result. 

Regards

Aamir
Hello Aamir,

the cause for the second image being not recognized is that it exhibits a three dimensional structure that cause variations in reflections and shadows when the lightning conditions change (e.g. direction of light, intensity etc).

Even though it seems natural to the human eye this is a very hard problem for image recognition algorithms to detect. So if you are using three dimensional objects as a target, you have to make sure it will be scanned from the same point of view and under the same lightning conditions as the target image was taken - otherwise it cannot be recognized.

On the positive side, we are currently hard at work to improve our image recognition engine to deal with situations like this, so this will definately improve in the future. However I cannot disclose any release timelines with you at this time.

In the meantime I would suggest to follow the guidelines stated above (ensuring same point of view and lightning conditions) or to use a two dimensional target object instead.

Best regards,
Nicolas

Hi guys, 


Please note the Wikitude SDK now supports object recognition and tracking.

You can find all details here


Best regards,

Paula 

Login or Signup to post a comment