Reliable skeleton tracking of 2 people facing each other during a normal interaction RRS feed

  • General discussion

  • Hi,

    Just finished the jumpstart session - what a remarkable device indeed...

    I'm looking to record conversing people using the skeleton feature capability. 

    Surely the face-2-face nature of the interaction introduces occlusions of joints if were to use a single device. To overcome this obstacle I intended to employ 2 different computers and sensors, each directed at a ~45 degrees horizontal angle to its respective target. 

    In the end I'd like to be able to visualize and process that unified interaction, that is to have the total data acquired by the 2 sensors synchronized in space and time, and to feed it to either a visualization tool or even better to KStudio for review and maybe ML. What's the best way to accomplish that?

    - As a side remark - could mirrors be employed in such a task to reduce the use to only 1 device?

    - Do the sensors manage to avoid cross interference? If not, does putting them back to back at 45 angle between them, or alternatively on top of each other prevent such an interference?

    Counting on you guys.


    Tuesday, December 2, 2014 4:18 PM

All replies

  • There are several threads that discuss multi-sensor scenarios. You will have to employ some type of networking of 2 systems and merge the 2 systems data into one coordinate space. This requires some type of calibration you will have to determine. This is not something provided by the Kinect SDK.

    For tracking body data, stability of the body tracking system will requires the head and both shoulders are visible to the camera.

    With Time of Flight cameras, interference is somewhat mitigated with the way the phase shifting happens, but there are points in time where things phase in and overlap causing missing data. Typically this will only last for a short period but can be cyclical. How or when it occurs would be non-deterministic.  

    Carmine Sirignano - MSFT

    Tuesday, December 2, 2014 7:39 PM