none
Skeleton Joint Quality and W RRS feed

  • Question

  • Hi All, 

    I've found some information on this topic from the "Skeleton Exiting View" at http://social.msdn.microsoft.com/Forums/en-US/kinectsdknuiapi/thread/b3885d99-67da-48ea-9801-ce58b7b7dc94

    However, I am now investigating skeleton tracking and I find that skeleton joint positions are sometimes pretty shaky in quality.  I get limbs / center of body jumping way out of position relative to the rest of the skeleton at times.   This seem to happen most frequently if I have two people near each other, or one person near some other 3D object (like a wall or chair or something). 

    "W" seems to be the value that gives you quality of a joint position, with a value of 1 being high confidence and a value of 0 being no confidence.  However, I'm seeing that no one is getting values beneath 1 here.  

    Has anyone else experienced this, and is there any SDK related solution (or other good solutions?) for measuring the quality of a joint in a particular frame to decide whether to use it or ignore it?  I am using the smooth skeleton tracking function. 

     

    Thanks


    Susan
    Monday, August 1, 2011 12:59 PM

Answers

  • There is no solution that will give you a very detailed granularity on joint position quality, but have you looked at JointTrackingState enumeration? This will tell you if each joint's position is explicitly tracked, inferred from positions of other joints (e.g.: when a joint is occluded by other joints) or not tracked at all.

    Also, there is the SkeletonQuality enumeration, which tells you at an even higher level if the top, bottom, left or right parts of overall skeleton have been clipped.

    Hope this helps,
    Eddy


    I'm here to help
    Tuesday, August 2, 2011 3:39 AM

All replies

  •  

    Oops - found another thread about the issue - "W" is not correctly documented and will always be 1 for now at "Why is Joint.Position.W always 1?"

    http://social.msdn.microsoft.com/Forums/en-US/kinectsdknuiapi/thread/e2dd4a85-5b6f-4efd-8375-12c75db37e54

    Still, anyone have an approach for measuring joint position quality?

     

    Thanks


    Susan
    Monday, August 1, 2011 1:01 PM
  • There is no solution that will give you a very detailed granularity on joint position quality, but have you looked at JointTrackingState enumeration? This will tell you if each joint's position is explicitly tracked, inferred from positions of other joints (e.g.: when a joint is occluded by other joints) or not tracked at all.

    Also, there is the SkeletonQuality enumeration, which tells you at an even higher level if the top, bottom, left or right parts of overall skeleton have been clipped.

    Hope this helps,
    Eddy


    I'm here to help
    Tuesday, August 2, 2011 3:39 AM
  •  

    Thanks Eddy, I will take a look into these - had not checked them out yet! 

     


    Susan
    Tuesday, August 2, 2011 12:08 PM