Skeletor mapping with the SDK RRS feed

  • General discussion

  • Hi Guys

    I've been out the development cycle for along while now and to be honest have no interest in returning in the short term.
    However as many of you know the Kinect SDK and Sensor has been used recently in the paranormal field.

    I am actually trying to without spending a huge amount of time trawling through the SDK Kit which I wouldn't be using anyway to figure out what actually maps a point for the classic stick figure... can someone point me in the direction of any info on how the SDK kit detects this points and if the other points in the figure are just based losely on the classic figure of it they are mapped out as well?

    Thanks in advance

    • Changed type Aaarrrahh Tuesday, June 12, 2018 10:15 AM
    Tuesday, June 12, 2018 10:04 AM

All replies

  • Skeleton mapping is based on machine learning. Can't tell you in detail but it tries to map the skeleton as best as it can on an image according to the data set it was trained against, which was probably extensive. The SDK provides confidence for the joints. Some of them will be inferred. E.g. if the shoulder joint is occluded but the elbow and collarbone are not, it will try to approximate since there's no way those two exist without the shoulder joint. In some cases though those approximations are very bad, so you check the confidence to validate them.

    Anything else you'll have to roll your own.

    Tuesday, June 12, 2018 12:59 PM
  • Here's the original paper:

    And accompanying video:

    Tuesday, June 12, 2018 3:15 PM