_gaq.push(['_trackPageview']); _gaq.push(['_trackPageLoadTime']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })();

Kinect Skeleton Test (new)

By On November 29, 2010 · 13 Comments · In KinectHacks.net News

CADET ( Center for Advances in Digital Entertainment Technologies) have been working to write an open source sdk for skeleton tracking. Here are some preliminary tests they have conducted and uploaded the video.

CADET are the people building Lyserg21 the open-source game for Kinect and they are a focussed team and we can expect some cool stuff from them in the future…

Related posts:

  1. First Visual Tests For Kinect Game Protoype
  2. 2D Skeleton Fun With Kinect
  3. Kinect Rubber Sheet
  4. First Kinect Head Tracking Test
  5. Kinect for the Web Through Javascript
 
If you enjoyed this article, please consider sharing it!
Facebook Twitter Delicious Digg

  • Pingback: Tweets that mention Kinect Skeleton Test (new) - KinectHacks.net -- Topsy.com

  • Pingback: Kinect for the Web Through Javascript – KinectHacks.net « Erik Champion

  • robvh2

    I recommend researching common physiological ratios to help pinpoint where joints would be located (including unseen ones like feet in particular). For instance, a typical human’s “wingspan” is roughly equal to their height, so a person with a 72″ reach should be roughly 6′ tall. You’ll find ratios to help approximate shoulder width, leg length, etc also. Nice work.

    • Cadetproject

      thx very much, you are right using human body background information about our topology helps a lot and is what we are working on. we are still in the process of researching the literature and papers of state of the art approaches to the problem and try to formulate our own solution on the basis of these…

  • http://joey1058.wordpress.com Joey1058

    Silly question here, I don’t know if these guys will see this. But I’m wondering if they could tap into an existing modelling program that uses skeleton based movements already, such as Miku Miku Dance? Although that program is freeware, I’m not sure if the tech could be considered open source.

    • Cadetproject

      sure thing, this will be of course a first demonstrator and benchmark for the sdk, but I didn’t know Miku Miku Dance before, that would be even more fun controlling mangas…

      • http://joey1058.wordpress.com Joey1058

        In that case, I’m glad I could help out! :-)

  • Pingback: Kinect人体姿态跟踪 | 增强视觉 | 计算机视觉 增强现实

  • Sapo

    looks really bad! i think feeding the depth video to eyesweb mocap would do a much better recognition already, you’d just need to do extract the missing depth infos

    • Cadetproject

      so do it if it is that easy, we would save a lot of resources and work..
      reconstructing a 2d skeleton (for the special case hands and feet are separatet and not merged or occluded by other body parts) is not at all challenging (which is more or less the second part of the video, which was done in less than an hour just for a students course so they have something to play).
      The first part is the actual interesting one which does already quite a good job for our first shot, Microsof needed almost 3 years developing a working and performant solution which is not performant enough (says Ubisoft so they developed their own) and which still has problems. who ever worked with computer vision knows that this is not a trivial problem. however if there is a solution already working which we’ve overseen we’ll be happy to know.
      A commercial solution (eductional license) for that problem with 12 tracking points costs 40000 Euro, the company didn’t want to tell me the price for the commercial license.
      So I guess we might not be the first or the best to deliver a good open source solution but I hope we can contribute code and ideas.
      best,
      Robert

      • Sapo

        oh sorry man didn’t mean to offend, i didn’t seen the video(i tought it was a image at first)now that i seen it… the first part is really bad, second part is good, better then eyesweb! maybe using a still from the second part of the video preview would give a better idea :) btw check out eyesweb, it’s free, made by a university and i think they have been working on 2d body tracking before microsoft even tought about kinect :) and as i said if you measure the depth for each joint you’d have decent 3d tracking..
        kinect xbox software is lot superior now indeed…
        i hope see cool stuff coming from you project, i did’t meant to be rude,just that green/blue image looks like really bad tracking :D

  • Dbags

    Add kalman filter state tracking ….

  • thedude

    and try realigning the monotonal basal lines…