Body measurements with kinect v2 RRS feed

  • Question

  • Hi, i'm very new to the kinect SDK and programming with it.

    I was wondering if it was possible to get accurate body measurements even if it is estimations.

    For those who knows the v1 and the v2.  I know that both the kinect use differents scanning technique(i.e Structure Light and Time of Flight).

    Which one would be more accurate? Also, what i'm searching for don't need to be in real time.

    Is it possible to get more data and that it would be simplier with a video ?

    Thank you!

    Thursday, April 27, 2017 8:08 PM

All replies

  • I can only talk about Kinect v2.

    The SDK by itself, cannot give you body measurements. You can use the skeleton API to query shoulder to shoulder or hips to knees etc.

    Depending on what kind of measurements you want it can get pretty bad. For example, if you want to query the waist of an overweight person, you can only get the portion that the skeleton gives you. The same estimation could also come from a slim person with the same joint placement. What I'm trying to say is, this way you cannot account for mass basically.

    Perhaps you could get some sort of estimation if you blend information from both the pixel domain and the camera domain. Like querying the pixel width of two joints(like the HipL and HipR) and correlate that with the pixel width that depth gives you for the user at the same region as the Hips joints from one side to the other and then convert it back to Camera Space. But I'm not sure if it would be reliable even as an estimation.

    Friday, April 28, 2017 8:23 AM
  • Hi! thank you for the answer.

    I've seen products like Styku who claims they can get body measurements more accurate then hand measurements with a kinect v2 and it seems like they use a turntable to scan the person.

    I know that microsoft has a software called 3DScan and 3Dbuilder but I don't really know how styku does it :( there must be some way possible with the kinect v2.

    Friday, April 28, 2017 2:19 PM
  • of course there is, but you have to write your own algorithm to analyze depth point cloud and/or color space analysis to get more accurate numbers.

    API exposed by Kinect.dll from the SDK does not contain a method to measure person height.

    Friday, April 28, 2017 2:34 PM
  • Thanks Jan for the answer!

    Can we extract infos on the depth point cloud or color space with the kinect SDK?

    Friday, April 28, 2017 3:10 PM
  • You typically use the X,Y cordinates you use on pixel space and then attach the depth value as Z. At least this is a way I know of.

    The SDK in itself cannot give you something as is. Like Jan said, you can probably do stuff but you'd have to write your own algorithm.

    Friday, April 28, 2017 3:15 PM