Log in or register to post comments

Scanning close up on ImageTargets

June 22, 2017 - 6:26am #1

I have 2 trackers in my scene: 1 is 1783 x 1273 px 300dpi (151 x 108 mm) the other is 5315 x 731 px 300 dpi (450 x 62 mm)

Both have a similar design but different enough that the ARCamera can tell the difference. Both have a QR code-type design on them, both trackers are 5 star rated on the Target Manager.

When I scan just the QR code design on the 1st tracker, it will augment. However if I try to scan just the QR code design on the 2nd tracker, it doesn't nothing until you move out more.

I've tried printing out the tracker in different sizes but to no avail. I have even swapped the QR codes around and it still works on the 1st tracker but never on the 2nd.

Unfortunately I can't post examples as it's for a client.

I have tried increasing and decreasing the amount of track points, changing the proportions of the tracker by adding height and dropping in the Stones tracker in the extra space, but nothing seems to work.

The only theory I have is that the 1st tracker has a lot of concentrated points on the QR code but the rest are quite sparsely spread out whereas on the second tracker, again there are a lot on the QR code but also a lot more generally elsewhere, many of them clustered.

Do Image Targets work on percentage? So a higher percentage of tracking points are seen on the 1st tracker but a lower percentage on the 2nd, despite the very similar amount of track points?

If anyone has any idea what I'm talking about or if you've had something similar happen, please help!

Thanks

 

Scanning close up on ImageTargets

June 23, 2017 - 2:47am #3

Hello,

Thank you for your detailed response.

Quote:

Are these the size values that you specified when uploading these targets to the Target Manager? Note that the Vuforia samples use meters as the default scene unit scale, so for #1 that would be 0.151m x 0.108m.

These are the dimensions of the JPEG that was uploaded but I set the width in the Target Manager to 600 for both, to make them easier to handle in the project. I tried re-uploading the 2nd one, setting the width to 5315 (which is the pixel width) but it didn't detect either.

Quote:

To better understand the behavior I need to better understand what you mean by "move out more". Here are some initial thoughts: In order for the target to be detected, all the keypoints must be in the camera's FOV Are you sure that detection is not happening? Depending upon your target size, augmentation placement and clipping plane settings, sometimes you have to pull further back for the augmentation to appear. The samples have a debug statement that indicates when a target is found which could be helpful in your evaluation.

When I say 'move out more' I mean move the device further from the tracker. In the Target Manager, there are a lot of track points on the QR design, so I have tried isolating that to see if the app augments from that, experimenting with the distance the device is from the tracker and printing it out different sizes. I find that I can be fairly close up to the QR design and it doesn't augment, yet if I pan along the tracker, it will augment once another part of the tracker is in the camera view.

There is nothing in the console that says it has recognised the tracker. Only when I reveal more of the tracker does it recognise it.

 

I have read your PM and followed your instructions

Many Thanks

Scanning close up on ImageTargets

June 22, 2017 - 2:19pm #2

Hello,

hollym wrote:

I have 2 trackers in my scene: 1 is 1783 x 1273 px 300dpi (151 x 108 mm) the other is 5315 x 731 px 300 dpi (450 x 62 mm)

Are these the size values that you specified when uploading these targets to the Target Manager? Note that the Vuforia samples use meters as the default scene unit scale, so for #1 that would be 0.151m x 0.108m.

hollym wrote:

When I scan just the QR code design on the 1st tracker, it will augment. However if I try to scan just the QR code design on the 2nd tracker, it doesn't nothing until you move out more.

To better understand the behavior I need to better understand what you mean by "move out more". Here are some initial thoughts:

  • In order for the target to be detected, all the keypoints must be in the camera's FOV
  • Are you sure that detection is not happening? Depending upon your target size, augmentation placement and clipping plane settings, sometimes you have to pull further back for the augmentation to appear. The samples have a debug statement that indicates when a target is found which could be helpful in your evaluation.

hollym wrote:

Do Image Targets work on percentage? So a higher percentage of tracking points are seen on the 1st tracker but a lower percentage on the 2nd, despite the very similar amount of track points?

Image target detection is not percentage based in that a certain percentage of feature points must be detected for a target to be recognized. Recognition is based upon the ability of the algorithm to extract feature points from the camera frame. Many of these influencing factors are environmental: lighting, quality of camera frame, motion blur, spectral reflections, etc. Additionally, 2D targets that are not printed to the same aspect ratio as the image uploaded to the Target Manager, or are not flat can negatively effect detection performance.

I'll PM you with more details, so please check your messages.

Thanks.

Log in or register to post comments