Start a new topic

Low light SLAM & Marker Triggering

We are evaluating transitioning to WIkitude for a new project - do you guys have any solutions for "low-lighting" situations?  I realize "out-of-the-box" wikitude SLAM and marker detection require fairly visible features to track (as does Vuforia -requiring over 20 LUX on iPhone 6 plus).

 

For Instance, another SDK company (to remain nameless) sent us a demo of extremely low lighting marker triggering (see attached picture) at less than 1 LUX using something like "contrast-limited adaptive histogram equalization", or CLAHE.


I actually reached out to some guys working on such a project and they referred us to that SDK (go figure).


Thanks,

M

 



We achieved it.  Let me know if you would like some assistance.


Low Light Marker Triggering: https://youtu.be/1vBxK-9HQ8c

Hi We also have a project low light level, less than 10 LUX. Anyone has any experience using CLAHE and Wikitude input plugins?
Hello Michael,

Unfortunately this is not something included in our samples so we cannot provide you with any code. If Wikitude does not recognize the picture then there is something wrong in your implementation. What you could do is maybe refer to the Input Plugins sample as the code provided there is the closest to what you wish to achieve.

Thanks
Eva

 

So we put some time into this approach - getting OpenCV to send the picture to Wikitude.  It seems Wikitude does not recognize the picture, though. 


Do you have any sample projects/code that we could compare with our efforts?


Thanks,

M

Hello Michael,

Unfortunately in low light conditions there is only so much that you can do to extract information from the camera frame. This could be a workaround for detecting targets under these conditions:
1. Use OpenCV carry out a histogram/CLAHE equalisation - this produces a very very noisy image but extracts quite considerably more detail from the camera feed
2. Feed the result from OpenCV into the input plugin of wikitude for the recognition
3. Based on your preference either display the unmodified or modified frame.

Alternatively, instead of histogram equalisation you could also try histogram stretching, though i assume the results should not differ greatly.

Thanks
Eva

 

Login or Signup to post a comment