Log in or register to post comments

Trying to learn how does Image Target work

April 10, 2019 - 3:23am #1

Hi

I am making an AR game which is trying to turn my college campus into game stage. So I attempted to use some object like sign as AR marker.

I took photos by my phone and clipped it to make AR marker

   

 

Both AR markers got five stars at Vuforia Target Manager. The blue one works perfect, almost got recognized immediately every time. However the red one is really hard to be recognized by Vuforia. Even when I am just trying to recognize the exactly same picture I uploaded as AR marker instead of the original sign in reality.

I read the tutorial (https://library.vuforia.com/articles/Solution/Optimizing-Target-Detection-and-Tracking-Stability.html) and clipping a smaller part of the original AR marker to make a new AR marker with higher density of feature points.

It works better than original AR marker, but still hard to be recognized by Vuforia. Sometimes it took me 3-5 mins to get the real sign recognized by Vuforia. But in every test trial, once it has been recognized, it starts to work well until the app shut down or restarted. 

 

It is frustrated to stand in front of the sign but AR just doesn't work while it is already five stars image target.

There are three things I am confused

1.Is there anything wrong in my marker? For example, do characters and words give features but it is actually bad for recognition? 

2.Does clipping huge marker into smaller marker with better feature distributed make better image target?

3.Why can it start to work well, once it has been recognized?

 

Could anyone give me some suggestion? Please

 

Thanks in advance,

Sherlore

AttachmentSize
Image icon EFx.jpg230.43 KB
Image icon EFx5.jpg76.47 KB
Image icon 擷取_2019_04_10_17_28_00_430.png239.63 KB
Image icon 擷取_2019_04_10_17_28_02_777.png262.38 KB
Image icon EC.jpg390.99 KB
Image icon EFx2.jpg144.39 KB

Trying to learn how does Image Target work

April 11, 2019 - 2:49pm #4

Hello,

Thanks @JohnGorv for supporting the community.

Here is an article for how to optimize a 2D planar target: https://library.vuforia.com/articles/Solution/Optimizing-Target-Detection-and-Tracking-Stability

Some additional tips:

  • The Target Manager's star rating provides a heuristic analysis of the given target, but does not necessarily represent the actual run time detection and/or tracking performance. It is to provide some level of feedback to help guide developers to optimize their targets.
  • The use of text as the primary source of image features for detection and tracking is not recommended. Text can lend itself to repetitive features from similar characters, and can also lead to false positive detections if more than one image contains the same text. Suggestion is to utilize a background image for the text that can be used as an Image Target, paying close attention to the optimization guidelines above. The areas of the background image where text will appear can be 'whited out' with image editing tools before being uploaded to the Vuforia Target Manager.

Thanks,

Vuforia Engine Support

Trying to learn how does Image Target work

April 11, 2019 - 2:51am #3

Hi Johnny,

 

Thanks a lot!! It sound exactly the problem I encountered. It helps a lot!!

I will update my marker with feature points distributed more balanced.

 

Thanks again,

Sherlore

Trying to learn how does Image Target work

April 10, 2019 - 3:38am #2

Hi Sherlore,

 

I've had similar problems to yours in the past and I'm pretty sure it's to do with the Target Image feature points being almost too densely populated in specific areas, especially when looking at something from a distance. For example, if you are trying to get the whole panel in as your target image you would need to stand back a bit, meaning that Vuforia might not be able to distinguish between feature points. But if you were to move closer then you wouldn't be able to see the whole target image.

Maybe try using just the text on the left hand side of the panel (the part in red)? 

This text seems like it has enough feature points and at suitable distance from each other to work.

That's just from my experience though, hope it helps a little bit :)

 

Johnny 

 

Log in or register to post comments