Log in or register to post comments

Photograph Image Target?

September 14, 2012 - 9:06pm #1

I am trying to use a photograph of a city panorama as an image target. When upload it to My Trackables page and try to convert it into an image target, I only get one, sometimes two stars of quality and the system doesn't let me use the photograph. There are, however, quite a few features that the tracking algorithm should lock onto, I think. I saw others using less sophisticated images as image targets. For example, on this site http://gavinmhackeling.com/?p=10, they use gallery images from afar and it works. On other examples, I saw a photograph of a single-colored shoe that worked well. How do people get these images to work as image targets? How do I make it get at least 3 stars so that I can use it? The Metaio SDK works quite well with my photographs, so it seems have better tracking capabilities, but it's extremely expensive, so I'd prefer to use Vuforia.

Thank you!

Photograph Image Target?

November 12, 2012 - 12:51am #9

Hi, thanks for this explanation of your use case, that's very interesting.

Although, as you said, Vuforia might not perform extremly well in outdoor scenes (due to visibility, lighting and weather conditions which impose much harder constraints), you may be interested in knowing that the upcoming release of the Vuforia SDK (version 2.0) will contain some very interesting new features like the "User Defined Targets" (UDT) and the "Cloud Recognition".

In particular, the UDT  feature will allow a user to build a target on the fly directly from images taken from the device camera at application runtime;

thinking of your case, you could imagine snapping a view of the city and instantly build a UDT out of it, which you can then use for tracking like any other Image Target; still, the outdoor view surely represents a challenging point, but it might be worth exploring this new feature as it would free you from building predefined targets.

Also, the Cloud Recognition feature will provide the ability to upload a huge amount of images to the Cloud (over the internet), thus removing the limitations of standard DataSets of Vuforia v1.5; that is another feature that could turn out somewhat useful for applications like yours.

 

Photograph Image Target?

November 11, 2012 - 8:56pm #8

Thank you for your replies. I am sorry for getting back to you this late. I was out of town with no Internet. You suggestions make sence, but they are not applicable to the project that I am trying to develop. What I am trying to do is to develop an app where you can go to a specific spot in a city and use your smart phone camera to point it at the city and view information about the landmarks in sight. I know this is tricky not only because of the contrast issues and camera quality issues, but also weather conditions. I was hoping this would work if the weather conditions are very similar to those when the image of the city skyline was taken. So there is no possibility of tinkering with the image as it's coming directly from the user's camera. I guess the technology isn't mature enough to do something like this yet. Thank you for your comments, though!

Photograph Image Target?

November 9, 2012 - 5:26pm #7

Yes do some testing. The target analysis for an image like this one is going to incorporate the sky region, which is effectively featureless for the purpose of detection and tracking. So the overall target rating is reduced. As Alessandro has suggested, you should try cropping the image to isolate the areas that present a good density and distribution of features. You can determine where these are by analyzing the TMS visualization of your current images - those yellow hatch marks indicate the location of features.

Also if your physical target is going to be large, you may need to compose the target as a MultiTarget, using multiple child targets, to ensure that a sufficient proportion of features are visible within the cameras field of view.

Photograph Image Target?

November 9, 2012 - 9:40am #6

FYI:

I made a small increase in contrast and brightness (brightness increased because the bottom part of the image is very dark and you loose detail there) and got 3 stars; 3 stars can work already pretty well.

Also, remember that star rating is an estimation, so with 2 - 3 stars it's already worth trying the target. 

Photograph Image Target?

November 9, 2012 - 9:26am #5

Hi, the fact that the tracking rate is 2 stars does not necessarily mean that the target will be bad;

Vuforia may be able to detect it anyway, and track it relatively well;

I think it is worth trying it anyway (you can use the ImageTargets sample to do a quick test), before coming to a definitive conclusion.

Then of course in your image I see two issues (where you can improve):

1 - contrast could be increased a little bit  (as you say there is already quite some contrast in the image, but sometime just adding a little bit more contrast may gain you an extra star,... worth trying it out)

2 - feature (details) distribution: the sky part of your image has little detail, so the features only cover the bottom half of your image; maybe slightly cropping the top part of the image so as to reduce a bit the sky (again not much, just a bit, but worth trying)

 

 

Photograph Image Target?

November 9, 2012 - 9:16am #4

Hi,

Thanks for replying! I just tried to create a trackable from this image (which is about what I need):

 

http://buddychai.com/Brighton/skyline.jpg

 

and I got only two stars on the Trackable creation website. To me, it looks like there is a decent amount of contrast in this image. I am almost certain this would be good enough for Metaio SDK to track, but Vuforia doesn't consider the image good enough.

 

At this point, there is no code involved. I need to get the trackable data file to be generated before I can try develop any virtual reality on top of it.

 

Thank you!

Photograph Image Target?

November 9, 2012 - 7:32am #3

Feel free to PM me w/ a link to your image and I can give you some advice.

Photograph Image Target?

November 8, 2012 - 10:55pm #2

can u show me your code? i can help you if u send me ur code

Log in or register to post comments