My first bit of computer vision code under the GPL:
It’s just a noddy method to try and identify different people, built on top of OpenCV’s face finder. It’s for the lirec project, as a potential way of identifying who is who for the scenarios involving the iCat.
by
Tags:
hi dave, what’s your experience with opencv’s face recognition, like how sensitive is it to light / scale / angles? Recently I’ve helped someone to do face recognition based art installation using a proprietary library and would like to swap it for opencv in the future.
oh, and how many faces do you plan to distinguish? Is iCat connected to a computer or does it have computer inside?
Hi Artem,
It seems pretty good with lighting and scale, but it’s sensitive to rotation. The only thing I can compare it to is some in house sony software I used to use which did the same job with a different implementation, and they are pretty similar iirc.
You could probably improve the rotation by adding more images to the training set, but I imagine it would result in lots more false positives.
I’m still trying to get a handle on how it needs to work, but I think we only need to distinguish between a few faces, maybe 10 or so at most. I think we can also cheat and get them to wear silly hats and glasses to help it out. The iCat is controlled from an external PC, we can also add extra machines to do the vision processing if need be – which is quite a luxury 🙂
nice. i hope this is going to work from fluxus soon :).
Oh, the one we used didn’t detect faces in glasses (even if they weren’t funny).
Leave a Reply to artm Cancel reply