Log in or register to post comments

Image Target Recognition Precision

January 15, 2019 - 2:15am #1

Hello,

I'm developing Hololens apps which use the ImageTarget functionality (but posted the thread in here because this seems to be more global).

I'm having an issue with the precision of Vuforia's recognition however. It seems that the algorithm allows detection when only some percentage of the feature points is detected. I came across this problem while manipulating QR codes. I indeed tested if the same code with only a part missing/altered could lead into a false positive, and it did (see the image linked to this thread, the altered one is detected as the original one).

Is there a way to tighten the conditions for the detected state to occur, at current release?

Then, if possible, an adjustable parameter accessible via the Unity editor in a future release would be a great feature for developpers who need precision  detection.

Thanks,

David

AttachmentSize
Image icon OriginalQRCode.jpg20.79 KB
Image icon AlteredQRCode.jpg17.24 KB

Image Target Recognition Precision

January 16, 2019 - 4:16pm #4

Hello,

There currently no APIs in Vuforia 8.0 SDK that allow for the modifying the detection precision, accuracy, etc.

Thanks,

Vuforia Engine Support

Image Target Recognition Precision

January 15, 2019 - 8:43am #3

Not necessarily, and especially in my case, the VuMark (which I have already worked with) comes with too much constraints for shapes, and it is a specific model. My thoughts were turned towards images in general, so Image Targets.

About the QRCodes, I used them to show my point (I saw the precision issue on another project, and quantified it more precisely using QRCodes which were easier to manipulate), I should have been more precise to avoid confusion, so sorry for that. 

Therefore, I still stand by my former question : is there a way to increase the detection precision for Image Targets?

Image Target Recognition Precision

January 15, 2019 - 5:15am #2

So you basically want the VUMark feature: https://library.vuforia.com/articles/Training/VuMark

Log in or register to post comments