View Full Version : [Discussion] Facial Recognition (OpenCV / Other)

03-09-2009, 01:28 PM

Your Face Recognition is implemented under OpenCV, right?

Yes, FR is a non-trivial problem. There are tools for implementing it in OpenCV, but you need to know what you're doing. It's not just "crack open the package, compile, and have a solution". OpenCV is a development api/library.

If someone starts a new thread in Programming, I'll respond with some ideas, however. I don't want to dump in Tony's thread here.

Okay, New thread! Let's talk Facial Recognition / Detection!

Thanks, Connor

03-09-2009, 01:47 PM
If I remember correctly (and I may not be) either Nuts and Volts or Servo did a series on openCV and facial recognition.

03-09-2009, 03:56 PM
Good ! i't such an interesting topic !

I am newbie about Facial recognition and openCV. From i have understood so far, OpenCV uses Haar training features to archieve this goal. I have tested the sample, seem working good, but however not so robust. If you rotate the head it lost it :p

It seems there are many ways. Skin detection is a way, but surely not enough, for a robust face detection. Perhaps some shape detection for eyes, ears and mouth would help. It would require some extraction of these head parts from the noisy cam picture..Dunno how to archieve it yet... Mouth may be tricky to detect has they change a lot depending on emotions if is reflecting, or if you are speaking.

Anybody has deeper experience on it ?

03-17-2009, 04:48 PM
maybe you could use the fact that the human face changes so much to detect it. People are always changing their faces, and so if you just detected an area that had constant change maybe that would work.

03-17-2009, 04:58 PM
There is a VB control out there, and I'd need to find it again (I actually registered it and was marginally disappointed, but it would work for most robotics applications), that is based around OpenCV that actually does an outstanding job of simply identifying a few faces from n# of random faces. It was pretty cheap, too, like <$100. I'll dig around my old bookmarks.