Everything you need is here: replace face by pen and eyes by phone. You could also try Local Binary Patterns which are faster and maybe easiest to train... Nevertheless, you need many true positive of your objects and false positive (images of everything that are neither pen or phone). You could also try Machine Learning algorithms included in OpenCV here, but you always need many images of your objects.
As an addition to this, if you are going to use a pen or phone in all possible orientations / scales, be prepared to get a gigantic data set and to actually train multiple models for different orientations. Or rotate image in steps and perform detection on that is also a way to go. Never try to include all orientations into a single model, it will create a poor model that works almost never (got this from experience)
Thank you for your reply but I did not understand how to use it is that I should create an xml file for each object (phone, pen)?
Asked: 2013-05-05 14:45:08 -0600
Seen: 424 times
Last updated: Jun 02 '13
Please stick to this forum for your problems about openCV. French isn't my mother tongue, so emails stating extra problems are NOT wanted on my work email. I have deleted your request by email and will continue to do so in the future. I keep my OpenCV activities restricted to this forum only.