none
Tracking passive infrared marker RRS feed

  • Question

  • Hi all,

    I'm developing an application for a project at the university and I've to get the coordinates of a passive infrared marker in the room. The marker is tracked with Kinect v2.

    Is it possibile? If yes, I don't really know how to get the exact coordinates, any help?

    Thank you,

    Elisa


    • Edited by ElisaN Tuesday, June 21, 2016 9:23 AM add a sentence
    Tuesday, June 21, 2016 9:22 AM

All replies

  • Hi,

    Kinect does not provide any tracking features other than body and face tracking. You can use color and depth information to build your own tracking algorithm for your particular object (the IR sensor). Then, you could use Kinect mapping functions to figure out which coordinates in the camera ("real 3D") space correspond to some pixels in the color (or depth) space.


    • Proposed as answer by jmmroldan Thursday, June 30, 2016 8:03 AM
    Tuesday, June 21, 2016 3:05 PM
  • Hi,

    Kinect does not provide any tracking features other than body and face tracking. You can use color and depth information to build your own tracking algorithm for your particular object (the IR sensor). Then, you could use Kinect mapping functions to figure out which coordinates in the camera ("real 3D") space correspond to some pixels in the color (or depth) space.


    Hi,

    Thanks for the response. I developed a method for doing that with the infrared frame. Now i need also de depth frame, but I'm wondering if I can read both infrared and depth frame at the same time (I need the same image for my calculations).

    Thank you,

    Elisa


    Thursday, June 23, 2016 9:05 AM
  • Yes, you can use the multiSourceFrameReader. Have a look at the CoordinateMapping SDK sample.
    Friday, June 24, 2016 8:18 AM
  • Yes, you can use the multiSourceFrameReader. Have a look at the CoordinateMapping SDK sample.
    Thanks a lot!
    Wednesday, June 29, 2016 7:14 PM
  • Hello! I was wondering if you've had success tracking your passive reflective markers. I've developed something very similar using the Kinect V2 in matlab, and I was wondering if you've been able to evaluate the noise or accuracy you were getting. Also - were you able to uniquely label and identify the markers through motion? 

    Thanks! 

    Tuesday, November 1, 2016 2:08 AM
  • Depends on whether there are other environmental factors that can create noise.

    The documentation says the IR frame exists for such purposes(as ElisaN did above), but keep in mind that your everyday clothes might function the same way passive markers work.This effect can happen due to fabrics,dyes or surfaces that are highly reflective or absorbent,when it comes to the band Kinect's IR transmits at.

    So assuming you've created a highly controlled sandbox and use non-reflective, non-absorbent clothing, you'll most likely have good accuracy/minimum noise.

    Tuesday, November 1, 2016 1:19 PM
  • Thanks for the response! I've done my best to control the environment and prevent the presence of other reflective surfaces, but I've found there is quite a bit of variability in the depth sensor for the pixels associated with the passive markers. How best did you segment out the passive markers from the IR frame, seeing as the direct IR reflective surfaces don't have a depth associated with it. 
    Wednesday, November 16, 2016 5:00 AM
  • You mean variability in the pixel coords that the depth in meters maps to? Or variability to the actual depth the marker is in?

    Variability in pixel coordinates is expected. Coordinate mapping works in a way that pixel coordinates are not expected to be the same in between frames for the same "object" nor are they the same for a difference depth to query at. If it was so , we would be able to cache the Lookup tables between frames. But it doesn't work like that.

    I haven't implemented a mocap or object tracking scenario before. Just familiar with the theory behind. Also due to work , I have lots of cases where reflection and absorption of the IR mess with our application quite visibly, so again very familiar with the theory behind it.

    As far as the segmenting of the marker, I think you need a threshold for the size of the region in pixels your marker occupies in order to track it. Single pixels won't work since variability of the coordinates takes over.


    Wednesday, November 16, 2016 9:08 AM