Face Detection and Tracking

[sixtdfson]

" A few years ago, two parents went out for dinner. A few hours later, the babysitter was calling to ask if she could cover up the clown statue in the kids' room,the father said,"Take the kids and get out of the house. We'll call thepolice, we don't have a clown statue." The "clown statue" is really a killer that escaped from jail.(true story)If you don't postthis letter on to 10 pagess tonight, the clown will be in your bed at 3:00 am with a chainsaw in his hand

[thiago24071983]

Interesting

[dbancajas]

Hi, Do you know the inner workings of the viola jones algorithm and the xml-structure? Like How do you normalize the resulting features calculated from the rectanlges? do you divide with the variance or the std deviation? In the paper it is the variance but some implementations on the internet use the std dev to normalize.

[piechulla1966]

What's "the paper"? Are we talking about this one?:

@INPROCEEDINGS{Viola01robustre­al-time, author = {Paul Viola and Michael Jones}, title = {Robust Real-time Object Detection}, booktitle = {International Journal of Computer Vision}, year = {2001}

}

[piechulla1966]

Ok, nobody else did, so I should provide you with the shortest possible answer: Since STD is the square root of VAR, your question boils down to "Is someValue / normalizingValue equivalent to someValue / sqrt(normalizingValue)". Nuff said?

[dbancajas]

yes. i am talking about the viola-jones paper. You didn't answer my question. You are full of rhetorical bullshit. I asked a simple question and didn't expect a convoluted answer. And FYI, I already found the answer, you normalize it using the STDDEV. Not the variance.

Is someValue / normalizingValue equivalent to someValue / sqrt(normalizingValue)" ??

It depends. if the normalizing value is 1 then it is equal? but if it is another number then it's not equal. nuff said?

[piechulla1966]

@dbancajas

Yes, I love rhetorical bullshit -- sometimes (it depends).

I do my thing, you do your thing.

I am not in this world to live up to your expectation.

And you are not in this world to live up to mine.

You are you and I am I.

And if by chance we find each other,

it's beautiful.

If not, it can't be helped.

[spider853]

1:05 nice

[threelegduck]

listen if you want to get more hits you need to have a trendy themesong - i can't emphasize this any more!

video was super cool ... i don't know much about this computer vision stuff, but does this open CV let you load in some kind of a 'training set' to track whatever you want?

[jk2l]

this kind of video is more a demonstration for researcher on how well does the algorithm work instead of showing cool thing.

in this case result and the quality of video is the only things matter

[piechulla1966]

It should be noted that the video is outdated. You can do this 10 times faster with the newest version of OpenCV.

[piechulla1966]

I don't care about hits. Yes, the classifier is a simulated neural network, which is "trained" with thousands of "positives" and thousands of "negatives".

[arielmoron]

do youhave a software?? it is open?

thanks

[piechulla1966]

This is the facedetect.c example of OpenCV with minor modifications. Check out the OpenCV library wiki.

Cheers, have fun!

[woow14610]

OPENCV harrcascading method :D

[piechulla1966]

aka "Cascade of Boosted Classifiers Based on Haar-like Features"

[swades1947]

cool... might work better if you could train the classifier on data with more noise or obstructions

[piechulla1966]

Another severe limitation (for 'real world applications') is: this classifier works on the 'frontal face' only. In practice, it's sometimes impossible to place the camera exactly in front of the subject.

[axewater]

cool stuff diud you program that yourself ?

[piechulla1966]

This is standard stuff. Just type "OpenCV" into your prefered search engine if you want to do it on your own!