I'd like to bump this thread because I'm having a similar problem with my Galaxy S5. 

I'm using Vuforia 5.0.6, Unity 4.6.9, JDK 1.8.60, ADK 24.4.1, Android 5.0,

and I created the scene using steps of sample project from

http://www.justapixel.co.uk/how-to-make-an-ar-app-in-5-minutes-with-unity-and-vuforia/

The project works perfectly in Unity Play Mode using Logtech c615 webcam.

*However, the target is never recognized by the Galaxy S5 camera.*

Any advice? 

PS: I've also tried setting the AR camera mode to maximum quality, switching targets, switching AR objects, tapping the phone screen to focus camera, moving camera and target to different angles, and holding the phone motionless for about 20 seconds.

Tried that also. Doest really work. Any other solution?

Not the autofocus;

go to QCARBehaviour component of your ARCamera and check the listbox of the Camera Device Mode Settings -> you can set it to MODE_OPTIMIZE_QUALITY

(nothing to do with the autofocus...)

I have set it to autofocus always. Is there anything else that can improve detection?

In Unity you can set the Camera Device Mode Settings of the ARCamera node, yes.

I am using unity3d. Are you talking on camera deice parameter?

No, there is nothing like that in the SDK; however you can perhaps try and use MODE_OPTIMIZE_QUALITY instead of MODE_DEFAULT in the camera initialization code (if you use Android, see the camera init() function in ImageTargets.cpp).

Threshold on number of points or something that compriises accuracy

What kind of  "threshold" do you have in mind ?

Is there a threshold that i can set?

I see. Actually I was thinking that the problem is with the banknotes because you confirmed that when printing the images (instead of using hte real banknotes) you observed a good behaviour, while the problems emerge with the real notes. 

But I acknoweldge the issue; at the moment I have no other ideas beyond what I have said already; I will make a few more tests on my side too, just in case.

Hi

I tested with bank notes. These are my results

1) Work ok with 50

2) Doesnt work with 10

The notes are new. I have even ironed them to check. The same notes get detected very easily in the web version. Honestly i dont see a problem with the notes

IN2ar behaves pretty well with all the notes that i have. 

User Defined Target does not work like the "predefined" targets; you just need to shoot at one banknote from close distance and trying to fit the camera view with the banknote, then you can move the device away/around and should be able to track the notes;

this is a test to see if those banknotes can be detected; I tested here with my Rs10 note and it did work pretty well (almost like in the case of the "printed" targets); 

if you cannot detect / track the banknotes with this UserDefinedTargets test, then it might mean that your banknotes are a bit ruinded or anyway very bad quality.

Doesnt work properly on 10 & 20. I have predefined targets. How do i change userdefined to work like predefined.

How do i use userdefined targets? Is there any samples?

Another test I did was also to use the UserDefinedTarget sample, pointing the device camera to the banknote and creating the target on-the-fly; 

this test also worked very well in my case; could you maybe try this test too ?

Doesnt work with scanning properly . Trying with camera image

I have tried scanning the note and also grabbing it from mobile camera. Both of them don't seem to work. Will try again today.

I actually get good results also with your original images (despite the 1-star only rating);

the point is that for some reasons, the real banknotes that you are using have some difference with the reference images (perhaps one good test is to scan your real banknotes and make Image targets using the scanned version... 

Printing the notes behave properly. On actual it takes a lot of Time and doesn't work smoothly. Can you mail me your images for 10,20,& 50. Will try with yours.

I have tested using a slight increase in contrast (getting 2-stars on Rs10 and Rs20, isntead of 1-star);

from the tests I have done it seems to behave relatively well (even on low-end device it typically detects the tracker within 2 - 3 seconds, and much faster on high-end devices);

the only thing I can think is that maybe the banknotes quality is perhaps not ideal; have you tried printing the banknotes images to paper (e.g. on a A4 paper sheet) and then testing with them ?

Just to see if it makes any difference, as here I cannot observe such a degraded performance... so not sure what else the issue could be.

Hi,

I have tried that also. Does work for 50,100.500 but doesnt work for 10 & 20. By optimizing the images, 20 seems to work but after a long time. Any other suggestions?

IN2ar works perfectly

 I've investigated a bit more this issue; what I have noticed is that using the 4-star enhanced images does not seem to improve the detection performances with respect to the 1-star images;

this is because the banknotes contain very special visual patterns which are used as a protection against "fake" money; these visual patterns can change significantly with the sampling rate (i.e. with the resolution of the image in pixels) ;

by downsampling the image to a lower resolution (e.g. width = 320 pixels)  as you did (as you applied the local contrast enhancement), you can actually alter those patterns significantly and (1) introduce new features that do not exist in the real banknote, or (2) remove existing features from the original banknote image.

So, it appears as a very special case;

what I would suggest is to try keeping the image resolution unaltered (i.e. do not resize the image), and simply increase (slightly) the global contrast of the image to make it appear less blurry; the idea is to be able to reach 2 stars (maybe 3 stars), but you don't need to push too much (no need  for the 4-stars). 

And then you can try again.

Hi,

• have you applied the tap-on-screen to trigger the autofocus on the low-end device ? NO. How should i do that?

yes, you should tap on the screen to refocus; if the device does not support the "continuous autofocus mode" (and low-end devices typically don't), the camera might get out of focus (i.e. not focussing properly on the actual target) and so your device will only see a "blurry" image, thus making the detection much much harder.

Meanwhile I'll see if I can test on HTC Desire S.... 

reply i

Ok, yes, please, send me the 4-star enhanced images if you can  (I would like to have both the 1-star and the 4-star versions for complete testing).

Just to check:

• have you tested against real banknotes or by printing the images (e.g. on an A4 paper) ? (BTW: I happen to have a real Rs10 banknote here and I'm testing on it as well)Yes I am testing with real bank note
• could you specify the low-end device model that you are using (so, if we have one here, we can test on the same device) HTC Desire S
• have you applied the tap-on-screen to trigger the autofocus on the low-end device ? NO. How should i do that?

Ok, yes, please, send me the 4-star enhanced images if you can  (I would like to have both the 1-star and the 4-star versions for complete testing).

Just to check:

• have you tested against real banknotes or by printing the images (e.g. on an A4 paper) ? (BTW: I happen to have a real Rs10 banknote here and I'm testing on it as well)
• could you specify the low-end device model that you are using (so, if we have one here, we can test on the same device)
• have you applied the tap-on-screen to trigger the autofocus on the low-end device ?

Hi,

I have followed the tutorial on those images https://developer.vuforia.com/resources/dev-guide/image-target-enhancement-tricks and got 4 star on rs 10 marker. If you want i can mail you those images. But the problem still exists

Got the images, thanks.

Here are the findings of my tests:

• The Rs10 and Rs20 banknotes only rate 1 star on the online Target Manager (not 4 stars); at least with those images that you sent (also, FYI I get 3 stars for the Rs50 and Rs500, and I get 2 stars for the Rs100)
• The main reason for Rs10 and Rs20 getting only 1 star is low contrast of the images; increasing the contrast should be able to raise the rating to at least 2 stars, maybe even to 3 stars (like for the Rs50 notes), with significantly better performances
• I tested on both high-end (Samsung Galaxy S3 with Android 4.0) and low-end devices (a two-year old Acer Liquid with 2.2) and in both cases I am able to detect all the banknotes (including Rs10 and Rs20), although on the low-end device I had to tap on the screen to make sure to trigger the right Focus for the camera (while the Samsung does not need this action, since it supports continuous autofocus)

 So, what I would suggest is:

• increase the local contrast of Rs10 and Rs20; see also this page on how to enhance local contrast: 
  • https://developer.vuforia.com/resources/dev-guide/natural-features-and-rating
• try again and make sure you tap on the screen to adjust the camera Focus so that it focuses on the target, when using low-end devices (which do not support continuous autofocus)

You can start with these steps, and then please report back if you need more help.

Just click on my name here on the right of the post and you will see a page with a Send Message button; then I will reply to that message...

How should i do that? Can i get your email address?

OK. could you PM me one of those images , so that I can test them and see what the issue could be?

Hi

I am testing in a relatively brighter enviroment. It doesnt work inrespective of the postion of the phone.

Hi, if the star rating is 4 (for all the targets) the detection and tracking should be quite good, even on low end devices:

still, some other factors that play a role are:

• lighting conditions (e.g. are you testing in a relatively dark environment ?)
• size of the targets; the banknotes are probably quite small and this could explain why detection is problematic if you stand a bit far; do you observe same results despite the distance from the targets ?

Star rating is 4.

Hi, what is the star rating of your targets ? (1 ... 5 stars ?)

If it is less than 4 - 5 stars, you probably have margin for improving their quality (tracking rating); for instance, it is often possible to gain an extra 1 or 2 stars by increasing the local contrast of the images.

With a higher rating, the detection andtracking performance should improve.

Please, have a look at these articles for more details on how to produce high quality image targets:

https://developer.vuforia.com/resources/dev-guide/image-targets

https://developer.vuforia.com/resources/dev-guide/natural-features-and-rating