[amendez]

How did you implement the hand detection? Since your using HSV color space, are your just doing color segmentation for a guessed range of Hue, Saturation, and Value for the color of skin? I have a few ideas on how to really improve the detection since I've worked on skin segmentation before

[shinao]

Oh nice, yes that's exactly what I'm doing, I take the middle of my hand, take the HSR point and increase it's range to get the whole hand. I'm listening :)

[amendez]

Nice! You can implement a real time skin segmentation algorithm using HSV segmented skin as training, and using random forests to do classification, it classifies in real-time with opencv!

Here is a video of my result from only training with one frame of a segmented HSV hand to some random videos: https://vimeo.com/114489035

Also, you can use a kinect or an occipital structure sensor to do segmentation from a certain depth, more hardware but less vision computation.

[amendez]

And using a depth sensor may be more robust for hand tracking and gesture recognition.

[shinao]

Thanks for the info, learnt something. What kind of project did you do for skin segmentation ?

[mendeza]

I was working on skin segmentation for a research project to it segment in real time for Google Glass! I never did the training with a huge database of ground truths. I want to eventually complete the training right to build a robust skin segmentation alg. I can link the github page(Opencv implementation only), but my code is really messy as I built as a POC in two days.

The google glass app was to use skin segmentation for gesture recognition.

[shinao]

Oh nice ! Maybe can be used for translating sign language in real time, that would be amazing, kinda hard to get the depth infos though.

Thanks for the answers and if you ever finish the alg you know where to post it :)

[mholmes680]

I don't understand anything you two just said, and it was awesome---this is why i come to HN. Nice work!

[option_greek]

What do you recommend for depth sensing ?

[Leszek]

We had a student once who used a face detector to first find the face area, and then extracted the skin colour from that region for the subsequent hand tracking.