Log in or register to post comments

Similar Image get recognized even if they are not created as a trackable

October 27, 2011 - 11:33pm #1

Hi All,
I have a few doubts regarding the Image targets QCAR application on android. I have created an image trackable through the QDevNet website. I have another image which differs only in color from the first image. I haven't created any trackable out of this. yet this image gets recognized when I run the application and track this. I have a third image which in quite different from 1 and 2, except that the x,y coordinates of the 3rd image is quite similar to 1 and 2. Even this gets recognized ( I haven't created a trackable for this one either). So here is my doubt :-
1. How many features(or points) are actually required for the recognition of 2 images that are different in color.
2. How can an image that is different in color but similar in x and y coordinates match with the trackable?

Regards,
Sarah

Re: Similar Image get recognized even if they are not created as

November 4, 2011 - 7:57am #6
Quote:

I want to know how much percentage of similarity is required between 2 or more similar trackables ( which differ in few features) to be matched as the same trackable.

I'm asking about this, but I'm not sure it's something we can quantify. This use case isn't really what the SDK is designed for, it expects to track off the same images you uploaded to the TMS.

Details of the tracking algorithm are not available, sorry!

- Kim

Re: Similar Image get recognized even if they are not created as

November 3, 2011 - 11:19pm #5

Thanks for the reply Kim.

As you said

Quote:

QCAR provides natural feature tracking. You can find some details in the Dev Guide: https://ar.qualcomm.at/qdevnet/developer_guide

I cant really find the exact details that i need. I want to know how much percentage of similarity is required between 2 or more similar trackables ( which differ in few features) to be matched as the same trackable. What image detection algorithm is used here? Could you please provide few details.

Quote:

I wouldn't use targets that are too similar in the same application. You may be able to track off all of them, but the tracker could have a hard time telling them apart (if you wanted different augmentations on each).

The application that I am working on is based on recognizing similar kind of images. I also need to use the same application for tracking real time images that are similar to the created trackable. So I would like to know how actually the image detection algorithm will help me in recognizing the images apart.

Regards,
Sarah

Re: Similar Image get recognized even if they are not created as

October 31, 2011 - 6:29pm #4

QCAR provides natural feature tracking. You can find some details in the Dev Guide: https://ar.qualcomm.at/qdevnet/developer_guide

I wouldn't use targets that are too similar in the same application. You may be able to track off all of them, but the tracker could have a hard time telling them apart (if you wanted different augmentations on each).

- Kim

Re: Similar Image get recognized even if they are not created as

October 30, 2011 - 11:31pm #3

Thanks for the reply Kim.

ksiva wrote:

The features (high-contrast corners) are what define a target, not the color scheme. Two images that are the same except for their coloring will most likely produce the same set of features. Compare the results in the TMS when you upload the images.

I was wondering which Image detection algorithm are you using. I would like to know more about it.

Quote:

I think I'd have to see the images to understand what you mean here, what do you mean by the "x,y coordinates"?

By x, y coordinates I mean that the way both the images are placed are quite same. Only a few features differ. So how does this happen?

Re: Similar Image get recognized even if they are not created as

October 29, 2011 - 11:36am #2

The features (high-contrast corners) are what define a target, not the color scheme. Two images that are the same except for their coloring will most likely produce the same set of features. Compare the results in the TMS when you upload the images.

Quote:

I have a third image which in quite different from 1 and 2, except that the x,y coordinates of the 3rd image is quite similar to 1 and 2.

I think I'd have to see the images to understand what you mean here, what do you mean by the "x,y coordinates"?

- Kim

Log in or register to post comments