Realtime image analysis
Author
I'm looking into technology for realtime image analysis, such as that used by the PS2's Eyetoy. Specifically, I'm interested in how I might go about tracking someone's hands as they move around (possibly wearing special gloves or something).
Does anyone have any technologies / references / articles / papers / websites / resources that I could look at? It'd be most appreciated.
Well, the Projet Odysée deals with visual recognition, so you could find it useful (although on the theory side rather than the real-time side).
They have published several papers on the subject: from their page.
They have published several papers on the subject: from their page.
You might want to check out some CS colleges for Computer Vision classes. They usualy contain lots of documentation about it. I have some for my class but it won't help you much since it's not in english. It also depends if you are just looking for algoritems or already made librarys/middleware.
The general problem of object tracking is a very difficult problem due to issues in background variation, occlusion, and the like. However, you would be on the right track to look into topics in computer vision related to tracking. In particular, you could use domain knowledge about how a hand is shaped to help you solve the more general problem for your specific domain. I found this paper particularly enlightening when I started my thesis work, perhaps it can help you handle adaptive backgrounds (if you need this) as well.
Magius
Magius
Author
Cheers, guys, I'll do some reading.
My journal has some info on what I'm thinking about trying to do, if you're curious. I'll try and take a look at the Eyetoy SDK tomorrow as well, see what they provide for you and what you do yourself...
I was definitely thinking that it would be a domain specific problem, not least because I'd want to try and calculate the hand's pose (i.e. position of the fingers) from the images. It'd be nice to be able to detect any number of hands (i.e. your friend sticks his hand into the shot) without the use of mocap-style gloves.
If already-made middleware exists then great, but pure algorithm type stuff is OK too. I'm more looking to see if this is plausable at the current level of technology, more than I plan to actually implement it.
My journal has some info on what I'm thinking about trying to do, if you're curious. I'll try and take a look at the Eyetoy SDK tomorrow as well, see what they provide for you and what you do yourself...
I was definitely thinking that it would be a domain specific problem, not least because I'd want to try and calculate the hand's pose (i.e. position of the fingers) from the images. It'd be nice to be able to detect any number of hands (i.e. your friend sticks his hand into the shot) without the use of mocap-style gloves.
If already-made middleware exists then great, but pure algorithm type stuff is OK too. I'm more looking to see if this is plausable at the current level of technology, more than I plan to actually implement it.
You may want to take a look at EyesWeb: http://www.eyesweb.org/. It's a free framework that among other thing lets you play around with tracking, gestrure recognition and various image processing. Also you can extend it by writing your own modules.
I just want to say, I only recently came across the EyeToy and I'm very impressed, especially considering the low specs of the PS2 (presumably why the games are so simple?).
Does the PS3 plan to support something similar with better games, and do other current/future consoles have anything similar?
Does the PS3 plan to support something similar with better games, and do other current/future consoles have anything similar?
Interesting. I'm currently studying for a PhD in the field of augmented reality. My specific area of research is using the hand as a pseudo-fiducial marker. A fiducial marker is type of marker that can be tracked and for which the position and orientation can be reliably estimated (rough definition). Currently I'm working with a sequence of very simple global image functions operating on streaming webcam input - specifically a skin colour filter, edge detector and contour filter plus a couple of others to improve image quality.
The next step after I can reliably extract and verify the contour of the hand will be determine the orientation. I am currently expecting to use a feature-tracker for this, probably line or texture based, but my research has not yet progressed that far.
I shall be writing a paper on my work so far to present at a conference in January. You're more than welcome to a copy when it's finished if you want.
Enigma
The next step after I can reliably extract and verify the contour of the hand will be determine the orientation. I am currently expecting to use a feature-tracker for this, probably line or texture based, but my research has not yet progressed that far.
I shall be writing a paper on my work so far to present at a conference in January. You're more than welcome to a copy when it's finished if you want.
Enigma
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
