Tracking hands through occlusion

A demonstration of a tracking algorithm which uses optical flow information to disambiguate between visually-similar objects. A two-stage dynamic model is used which allows tracking with only a small number of particles in the particle filter (25).

Category:

Science & Technology

Tags:

• Computer
• tracking
• optical
• flow
• human-computer
• interface

[mvharish14988]

Thank you for replying so promptly!

Ah, so let me clarify, you use skin color based segmentation as well as sparse optical flow? Please correct me if I speak wrongly.

[MatejKristan]

@mvharish14988 There is no explicit segmentation. The color histogram (adaptive) is used to model the target's color (but not necessarily skin color). The whole thing is implemented in a particle filter, so you don't require explicit segmentation.

And you are right that I use the sparse optical flow. If you're interested, there are some additional videos on my homepage.

[mvharish14988]

@MatejKristan Yes, I looked through your website. I did not realize you were a Professor. =)

And btw, that's a nice article on chicken-egg problem...

[MatejKristan]

@mvharish14988 Thanks. We can learn a lot form the chicken-egg problems :). But I'm not a "full" professor yet, and probably that will take a while :).

[mvharish14988]

Cool, but how does it work if you change the shape of your hand?

[MatejKristan]

@mvharish14988 It does not take into account the shape features, so it's robust to changes in shape. It only relies on color (histograms) and local motion (flow) features.

[mvharish14988]

@MatejKristan I think the article is the best resource I have found so far on why Expectation Max. algorithm is so important, and I would definitely recommend it for anyone wanting to understand EM. And good luck for the Professorship, I can see it happening! :D

[esurior]

Never mind, I just noticed the reference to your website :)

[esurior]

Impressive. Did you publish your work or based it on some specific papers?