none
Unwanted depth pixels RRS feed

  • General discussion

  • Hi,

    I'm working with the new Kinect for Windows v2 sensor.

    I have taken depth images from hanging objects (balls) and their sides blur with pixels forming something like a cone. So basically it takes up pixels where it shouldn't.  I would like to know how to solve this problem and remove the unwanted pixels.

    Thanks for your help and time :)


    Tuesday, September 30, 2014 9:40 AM

All replies

  • Do you have a screenshot of what you are seeing? As with any image filtering, you will have to do some post processing on the data from withing your code. You can apply an averaging filter that should remove large changes in frame data, but if the ball is moving producing a motion blur, then that is going to be a more difficult scenario.

    Carmine Sirignano - MSFT

    Monday, October 6, 2014 4:55 PM
  • I can't insert images.

    I have problem with my e-mail address.

    The photos (3 ball , in front and side view) can be found here:

    mega.co.nz/#F!pAtllaiI!dQJblI0fBQolijFHbPKLyg

    What do you suggest?

    Tuesday, October 7, 2014 7:09 AM
  • this was a big problem for me,

    several approaches to tackle it, the current approach i'm using is to compare the

    distance between a pixel and it's neighbor; if the gap around it is bigger than 3x it's size then I exclude the pixel.

    The pixel size is, well the size of area that the pixel represents. Pixels further back are larger than closer ones.

    It's best to also run a smoothing algorithm before doing this check, otherwise you'll get quite a few false positives

    Doing this has removed a majority of the bad data.  Now I'm still working on the edges and handling things that get indented because they seem to absorb the infrared differently. Sigh, so much to do.

    Another option is to add the difference between a depth pixel and it's neighbors together, if a neighbor is missing then just use the average of the neighbors that are there; then filter out pixels that have a high difference, as those pixels are scattered anyways.


    • Edited by Jacob the Geek Wednesday, May 20, 2015 10:14 PM added another method
    Wednesday, May 20, 2015 12:42 AM