About image recognition
Opened this issue · 5 comments
Can I use ezar to do image recognition, and show 3D model with three.js?
Image recognition is not yet available but in the works. And yes you'll be able to use three.js features with. For example here is an experimental project I created with ezAR videoOverlay plugin and three.js https://github.com/ezartech/pokemon-go-experiment.
I'm assuming you are looking to identify an object and replace it with a virtual object or something similar? Let me know more specifically as the more feedback the better I can tune a solution to fit.
We already used the videoOverlay plugin with three.
That is awesome.
Yes, we have a game like Pokemon go, but we need summon character from scan a card, so that need image recognition or text recognition .
I found opencv and tesseract-ocr, but I can't use them with ezar videoOverlay.
Thanks.
Send me an email to discuss a possible opencv-based solution, wayne@ezartech.com.
I already send you an email.
Thanks.
For image recognition, I suggest to use JSFeat. It can do image matching and recognize object location.
Jsfeat examples are on webcam streaming only. I was able to change the target to be an image. The example is attached below. Link to jsfeat: https://inspirit.github.io/jsfeat/
Way forward is to let jsfeat read from the ezAR camera.
Example of image matching and object location.
sample_orb.zip
All the required javascript files are in the below attachment.
js.zip
Replace the html image1 with your image of a physical object such as magazine cover or your photo :)
Run the example on a web browser and don't forget to press on train_pattern button on top left.