Fiducial marker displacement tracking using CUDA

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,
I’m currently tracking Aruco markers using Opencv.
I need to get displacement at millimeter or sub millimeter level.

Aruco markers can achieve the goal but no CUDA is used.
The fps is somewhat slow as opencv is using cpu cores.

Does the Aruco libray been implemented with CUDA ?
Is there another fiducial markers library that use CUDA and can track sub millimeter displacement ?

Thanks.

2

Just a google search turns up: GitHub - nbergst/cuda_aruco: Aruco library with cuda optimizations

If you need accurate tracking, you could have a coarse detection and then use CUDA to finetune. But it seems you are looking for a usable library instead of implementing it yourself.

3

Thanks Curefab,
I never came accross that repo during my extensive google search.
Yes we could go the way of CUDA for finetuning, but why bother reinventing the wheel if it’s already been done ?

4

If you have the necessary domain knowledge, possible motivations could be:

(1) You need a faster solution than what is offered by the library
(2) You need a more accurate solution than what is offered by the library
(3) You need functionality not offered by the library

The recommendation to use existing libraries instead of “rolling your own” is usually sound advice, but not always.

I have been involved with mathematical libraries, and in that space I still see a lot of porting of old Fortran codes from the 1970s or C codes from the 1990s into modern programming environments although the state of the art for both hardware and software have changed drastically since then. Consequently the design decisions made fifty years ago are often not the ones we would make today, with resulting impact on overall quality regarding the aspects enumerated above.