Log in or register to post comments

Vuforia Fusion: inconsistent precision in tracking pose

November 13, 2018 - 1:05am #1

Hello,

I was testing the new Vuforia Fusion Tracking Pose functionality and I found out that it seems like to work very good on some occasions while in others the same tablets (Samsung Tab S3 / S4) with the same build works poorly (the image target or the object target is not well detected and the extended tracking is not precise e.g. the extended model moves with the tablet).

It is very strange beacause the precision can get significantly better from a build to the next or even just waiting for an hour withouth touching the tablet.

I tried with different lights conditions, I tried to restart the tablet, deleting the builds, build from another computer, made an empty project, but OR everyting works poorly OR everyting works good. When it starts to work good, it keeps its good precision until like the next day. Then the cycle restarts.

I'm out of ideas, do someone know what is happening?

Thank you in advance and have a nice day!

Vuforia Fusion: inconsistent precision in tracking pose

November 27, 2018 - 10:08am #4

Hello @fcs1,

Per Vuforia Fusion, platform CV enablers (ARKit, ARCore) will be prioritized above Vuforia VIO as we understand that platform CV solutions have a distinct advantage over our tech.

However, we continue to improve upon Vuforia VIO and will release updates soon to improve the quality of the experience.

Thanks,

Vuforia Engine Support

Vuforia Fusion: inconsistent precision in tracking pose

November 22, 2018 - 12:50am #3

medabit wrote:

Hello,

Due to the nature of the technology, VIO technology (ARKit, ARCore, Vuforia VIO) may vary in the quality of the 6DoF tracking experience it is able to provide in different environments. This article includes some information for an ideal Ground Plane environment: https://library.vuforia.com/articles/Training/ground-plane-guide.html

Specifically, the following is relevant to this issue: Ground Plane utilizes images from the device camera to understand the basic geometry of your environment. Therefore it is necessary to stage Ground Plane experiences in settings that enable the camera to accurately capture the details of the surroundings. 

Recommendations:

  • Stable lighting conditions
  • Moderate lighting - not too bright or too dark
  • Avoid glare and dark shadows
  • Surface details help to reveal geometry and improve accuracy and performance

Doing our own experiments, we've found that in areas with low detail surfaces we were able to improve the experience by taking additional time to scan the environment before performing the first hit test. While not guaranteed to work, we were able to perform 6DoF/positional tracking in a wide variety of environments such as low detail carpet, tiles, and pavement.

The basic thought is that the more information the system has from environment scanning (floors, table tops, etc.), the more likely you are to gain a 6DoF pose. You can also use our tracking quality APIs to help better drive feedback to the user: https://library.vuforia.com/content/vuforia-library/en/articles/Solution/tracking-state.html

The Ground Plane feature will continue to improve with future Vuforia releases. APIs that provide information on the tracking quality will continue to be refactored for harmonization with platform CV providers (ARKit, ARCore), and improved feedback.

Thanks,

Vuforia Engine Support

Thank you for your very detailed answer!

Unfortunately, even testing with same light/surface conditions give me different results at different times, and I don't get why.

Seems to me that ARCore is still more accurate than Vuforia Fusion (which should use ARCore if I didn't misunderstand).

Vuforia Fusion: inconsistent precision in tracking pose

November 13, 2018 - 8:57am #2

Hello,

Due to the nature of the technology, VIO technology (ARKit, ARCore, Vuforia VIO) may vary in the quality of the 6DoF tracking experience it is able to provide in different environments. This article includes some information for an ideal Ground Plane environment: https://library.vuforia.com/articles/Training/ground-plane-guide.html

Specifically, the following is relevant to this issue: Ground Plane utilizes images from the device camera to understand the basic geometry of your environment. Therefore it is necessary to stage Ground Plane experiences in settings that enable the camera to accurately capture the details of the surroundings. 

Recommendations:

  • Stable lighting conditions
  • Moderate lighting - not too bright or too dark
  • Avoid glare and dark shadows
  • Surface details help to reveal geometry and improve accuracy and performance

Doing our own experiments, we've found that in areas with low detail surfaces we were able to improve the experience by taking additional time to scan the environment before performing the first hit test. While not guaranteed to work, we were able to perform 6DoF/positional tracking in a wide variety of environments such as low detail carpet, tiles, and pavement.

The basic thought is that the more information the system has from environment scanning (floors, table tops, etc.), the more likely you are to gain a 6DoF pose. You can also use our tracking quality APIs to help better drive feedback to the user: https://library.vuforia.com/content/vuforia-library/en/articles/Solution/tracking-state.html

The Ground Plane feature will continue to improve with future Vuforia releases. APIs that provide information on the tracking quality will continue to be refactored for harmonization with platform CV providers (ARKit, ARCore), and improved feedback.

Thanks,

Vuforia Engine Support

Log in or register to post comments