Start a new topic

3d object recognition

we are trying to create wto file for a 3D printed model of a kiosk using the wikitude studio. Some sample photos of a kiosk are attached.  We have seen that 3d object detection is more succesful when we are looking from the front side of the object. From the left side, the right side or from the rear side of the object, object detection is not running as good as we wish. what is your recomendation.


Hi KADİR,


We've analyzed the images used and realized that the stickers on all the faces are exactly the same. So the algorithm mixes up the different sides.


One way would be to have different stickers on different sides.


Should you need anything further, just let me know anytime.


Greetings

Nicola

Hi,

Each time we upload the same images, we get a different wto object after the process at your side. Here is the link of the images we use for target object creation.

 

Password: 1q2w3e4r

https://idrive.innova.com.tr/index.php/s/7Pj0LdzUiuF12ts/authenticate

 

 

An other point is that sometimes the faces of the 3D object are mixed with other faces of the same 3D object(for example front face of the object seems to be extended with left or right face of the object) when we open the wto object in the wikitude studio, and some faces are rotated 90 degrees.

 

The wto objects are under user cosar.kadir@gmail.com account:

 

2901_vnd210_3  is the Project name.

 

vnd_210_4

vnd_210_3

vnd210_2  are the object target names.

Hi,


The side and back don't have much structure or contrast - this might be the main reason. Our object recognition currently is working based on feature points that can be extracted within the object (so any difference in color, struture, etc. helps for the extraction). 


From what I see in the images, if the object only has this grey wall color, the object Recognition won't work properly. If you can, please add structure.


Thx and greetings

Nicola

Hi,

I attached the files.

rar

Hi,


Can you please resend the pictures? I couldn't find them attached.


Thx and greetings

Nicola

Login or Signup to post a comment