Topic locked

Best Practices for markers far away

January 15, 2013 - 6:25am #1

I have to create a marker that is 5.5 inches by 4.5 inches in size and I need to be able to pick up the marker 15 feet away if possible.  

What's the best way to go about creating a proper marker for that distance and size target?  Is it doable?

Are there are considerations with the api (settings in unity) to help with this?

Here's a pic of the target:

http://content.screencast.com/users/neoRiley/folders/Snagit/media/26666092-688b-426c-89cc-b647cafccdae/2013-01-15_08-23-50.png

 

Thanks for your help,

 

John

Best Practices for markers far away

July 27, 2014 - 10:33pm #17

This may depend on various factors, such as the Focus mode used (is the camera image sharp vs blurry), the lighing conditions, the Image Target star rating (e.g. you could try with a different target to see if you observe any difference) and other factors; this is all summarized here, with tips and how-tos:

https://developer.vuforia.com/forum/faq/how-can-i-improve-detection-and-tracking-stability

 

Best Practices for markers far away

July 27, 2014 - 8:15pm #16

I'm having a similar issue as the OP and have yet to get a response on my own thread in the past month and a half, which is why I thougth that I should ask here. I'm using the Books sample as a base for my own app. I have several floor mats, about 2feet by 3feet in size,  with different designs on them. I want to have Vuforia recognize them via the Cloud Database and ImageReco and overlay a product description on it just like the Books sample does. I use the original high resolution artwork of the mats as the image marker sample I upload to the cloud database.  Still, the app only recognises the matt when I'm no more then 2 feet away. After that it does not. Even with very good light it still only does about 2 feet (70cm).
However, once the mat has been recognized, the app does track the image for a pretty significant distnace (up to 5m). I'm desperately trying to find a way to increase the recognision distance of the marker so that it could be recognized by the app at about 3-5 meter distance. I have went through many threads on here ans tried several approaches but I'm yet to find a solution. I've read that people are achieving 3-5 meter recognition distance pretty easily with much smaller markers, so I'm wondering why mine only work at 2 feet away or less.  Please help

Best Practices for markers far away

January 22, 2013 - 11:18am #15

You should post to the wish list if you're interested in having such a gallery available - https://developer.vuforia.com/forum/general-discussion/wish-list

Describe some of the cases that you'd like to see supported. 

There is such a broad diversity of viable target images that we haven't needed to provide explicit examples, but I can see where this may be helpful for outlying cases. 

Best Practices for markers far away

January 22, 2013 - 10:47am #14

Thank you neoRiley for taking the time out to reply. I will have a similar requirement in the near future and will post my results too.

Best Practices for markers far away

January 22, 2013 - 8:22am #13

btw, the front facing camera feature of 2.0 works perfectly ;)

Best Practices for markers far away

January 22, 2013 - 8:21am #12

Thanks very much for the information David!  I hadn't realized you could do that with 2.0 - I'll definitely be looking at that ;)

and thanks for the info on your measurements - at least my findings were consistent with your rule of thumb (5ft approx).  

One thing that I wish we had access too is a gallery of markers that satisfied different needs so that we have a clearer understanding of what we're trying to accomplish.  Just wondering if Qualcomm has considered doing something like this?

Best Practices for markers far away

January 22, 2013 - 7:38am #11

In my experience, the maximum reliable detection and tracking range for good targets is typically about 10x the diagonal length of the target face. You can extend that a bit with deliberate design refinements and testing, but there is a limit. The diagonal of your target is about 6.4" and so 15' is ~ 28 times that length.

Also our 2.0 SDK is capable of generating trackables at runtime. This is demonstrated in our User Defined Targets sample. You can use this for testing purposes, to evaluate targets without needing to upload each of them to the Target Manager. 

Best Practices for markers far away

January 22, 2013 - 5:55am #10

Thanks a lot neoRiley for sharing your current findings in such great detail.

The results of your experience will help many others for sure (and by the way, very interesting ).

Best Practices for markers far away

January 22, 2013 - 5:32am #9

Well, the findings were not much different than what was discussed.  I was using an iPhone5 Rear camera for the demo.  I tried both framemarkers and image markers.

Frame markers worked well actually.  However, when you loose any part of the frame from view, you loose the target.  But I was able to print out a great quality framemarker (which is key, get it as crisp as possible) and pick it up from 8ft away.  The marker was printed on 8.5"x11" sheet of paper and was actually 4x4 or so square in the middle of the paper.  The bigger the print out, the better obviously.

Ultimately, I ended up using the image markers that come with vuforia since this was a proof of concept for a client.  I printed them out at 5x4 and put them on these fabricated plastic "L" shaped holders (it's for a game  - http://rockonflash.com/infrared5/clients/MarkerTests.JPG ).  The framemarkers were used on the bottom.  You shoot the main imagemarker that knocks it over, which then sees the framemarker to trigger an explosion.

I had success from about 5-6ft away consistently.  Anything above that was not very reliable.

We went back to the client and reported the same findings stating that if we moved forward, we'd have to spend a decent amount of time figuring out how to reach 15' reliably.  At this point, I don't know that it can be done with markers that are 5x4, in fact, I'm doubtful.

So, the bottom line is:

1.  Clear as you can print outs

2.  Large as you can print outs

3.  Make sure the markers don't curl - print on card stock if you have it.

Also, for another POC i'm working on right now (same client), I was able to use a semi transparent business card as a marker with an iPad front facing camera (relatively close, 1'-2' away max) which I thought was extremely cool.  We're using it to fly a ship on screen and using virtual buttons.  To see where the buttons are, we needed to use semi transparent materials and that actually worked.

 

hope that helps!

John

Best Practices for markers far away

January 22, 2013 - 12:22am #8

indeed, it would be nice to see if neoRiley has got some results already...

Best Practices for markers far away

January 21, 2013 - 5:16pm #7

Hey,

Wondering if you've had success? Would love to know some specifics - kind of camera, distances you've been succesful with.

Glad that I'm useful. Not

January 15, 2013 - 9:21am #6

Glad that I'm useful.

Not sure if a frame marker would behave better in this case;

in the case you describe the problem comes from the small size at very high distance anyway, I presume you would likely face the same issue (with roughly the same results) using Frame Markers.

 

it DOES help tremendously -

January 15, 2013 - 8:47am #5

it DOES help tremendously - you've at least confirmed my findings and that I'm headed at least in the right direction :)

Thanks for the info on the star rating.  I had thought that was the case, but it's nice to hear it from someone with knowledge of the situation

Thanks again, and I'll post my findings when I have some results

John

 

PS> would a frame marker perform better for any reason?

Hi, as you corretly say

January 15, 2013 - 7:36am #4

Hi, as you corretly say ( "alot of triangles in the marker and it becomes a blurred mush of lines" ), at such high distance the target (as captured by the device camera) will not be sharp and detailed enough (this is indeed why I was making the comparison with a 1-inch target at 3-feet distance, as it highlights exactly this problem).

So, indeed, it's a really challenging use case... (although you can still make quick test and verify if it is likely to be feasible or not)

Concerning your second question:

      So, what Im wondering, given this type of scenario, wouldn't it make sense to create a marker that has more general features that from even a large distance away

One thing to keep in mind is that the number of features has quite an impact on the tracking rating, and thus on the ability to detect and track the target;

on the other hand, however, you don't necessarily need to have a 5-star rating to make a target detectable; so, you could create a target which has less features, i.e. more sparse features, (and which may result for instance in a "3-star rating", for example) but which looks a bit "clearer from high distance (because it will have less triangles, as you say).

In other words, it's a bt like trying to find a trade-off between having a highly rated feature-rich image vs. having a few less features which look sharper at higher distances;

I don't know to what extent such trad-off can be found, but this is maybe something that is worth at least a quick test.

I hope this helps.

 

Thanks very much Alessandro -

January 15, 2013 - 7:19am #3

Thanks very much Alessandro - I appreciate your time and help

I adjusted the far plane, so thanks for the tip on that ;)

 

One thing I'm wondering about is the marker itself.  I created it using an online tool and of course, when I imported it to vuforia it got a 5 star rating as a good marker.  The problem I'm concerned with is that it's so complex that how can it possibly be viewable by the camera at that distance?  ie: less pixels to work with, alot of triangles in the marker and it becomes a blurred mush of lines.

So, what Im wondering, given this type of scenario, wouldn't it make sense to create a marker that has more general features that from even a large distance away, the API would be able to pick it out?  like for instance, a big biohazzard logo on the same size area.  Im guessing I don't know enough about how the algo's work.

Thanks again

Hi neoRiley, first

January 15, 2013 - 6:58am #2

Hi neoRiley,

first consideration: your use case sounds a bit challenging;

in terms of proportions in the camera Field of View, a target of 5 inches at a 15-feet distance is roughly comparable (using a linear approximation) to a target of 1 inch (i.e. very small) at a 3-feet (about 1 meter) distance;

in addition, the high distance may introduce other disturbing elements, e.g. lighting conditions not uniform in the room, insufficiently sharp focus of the device camera lens;

I'm not entirely sure about how well this will perform, but you could do a quick test with ImageTargets by printing the chips/stones targets with a slightly smaller scale (on a regular A4 paper they measure approximately 9 - 10 inches, so you can shrink by 40-50% or so).

On the API and Unity side, one thing to check is:

- ARCamera (Unity editor) => make sure to adjust the near and Far distance planes s to accomodate this case (otherwise you risk to get your 3D models clipped by the Far distance plane); 

I hope this helps.

Topic locked