Log in or register to post comments

Which is more robust, model or multi target tracking?

July 19, 2019 - 6:53pm #1

I'm looking to design a tracking object that is to remain stationary but able to be viewed at different angles. Which is the more robust tracking method, the use of a model target or a multi-target? My initial inclination was to design a sophisticated multi-target using a polyhedron and multiple QR codes. This looks non-trivial, however, and so the multi-target would likely be a cuboid with multiple QR codes. Would this be more or less robust than some other shape that could be tracked as a model? Thanks!

Which is more robust, model or multi target tracking?

July 22, 2019 - 11:43am #2

Hello,

Authoring your shape as a Model Target will mean that the Engine algorithm will use the object's geometry for detection. If your polyhedron is symmetrical in all orientations, this may cause confusion for the detector as to the correct orientation of the object upon detection. Thus, the "up" orientation could be detected as "down".

Agree that Multi-Targets may be able to work for this use case, but this will require hand editing of the device database's XML file to modify it such that the polyhedron's "faces" are properly represented in 3D space with respect to object's origin. Here is a high-level overview of this process:

  1. Create a new device database in the Target Manager
  2. Upload all images to be included in the MultiTarget as single image targets (with appropriate size values)
  3. Select all images in the Target Manager, and download the device database (non-Unity Editor packaging)
  4. Hand edit XML file to create the custom MultiTarget: https://library.vuforia.com/content/vuforia-library/en/articles/Solution/How-To-Modify-the-Dataset-Configuration-XML-File-Outside-the-Target-Manager.html#how-to-modify-multi-target

Lastly, perhaps you should consider the Object Target's feature: https://library.vuforia.com/articles/Training/Object-Recognition. This feature uses the object's surface details (i.e. QR Codes) to create a SLAM map of the object via the Object Scanner App, which is authored into a device database via the Target Manager.

My only concerns for Multi-Target and Object Targets is whether the QR codes provide enough district features for them to be accurately discriminated. This could easily be tested with Image Targets by creating unique targets for each QR code, and confirming that Vuforia Engine is able to correctly and uniquely identify each as unique Image Targets.

Thanks,

Vuforia Engine Support

 

 

Log in or register to post comments