none
manual exposure correction

    Question

  • hi,

    as many developers i need to correct the automatic exposure correction of the kinect v2. i read in the forum, that there will be no way to deactivate this feature (is this still true?). thats why i want to manually undo the automatic exposure correction, using the gain, exposuretime and gamma which are given in the colorcamerasettings.

    can somebody (microsoft developers?) tell me how these parameters are used to compute the RGB data? there should be a possibility for a reverse operation. In the image below you can see the exposureTime of 98 frames of a testscene (upper subimage) and the luminance values at one pixel in every frame. As u can see it looks realy similar but i couldn't figure out a solution that works on other testscenes as well.

    Tuesday, September 15, 2015 6:46 AM

All replies

  • You can find more info on this topic in this older thread, including some very generously shared code: https://social.msdn.microsoft.com/Forums/en-US/2a6ba4d8-8c78-4c43-9345-d4136382d629/auto-exposure-compensation?forum=kinectv2sdk

    Note that once bright or dark values get clipped when over/under exposed and in low exposure mode the color switches to 15 fps.
    These lost values/frames can not be recovered by any image algorithm since they simply are not there.

    Brekel

    Tuesday, September 15, 2015 9:26 AM
    Moderator
  • Thank you for your answer. I already read the thread from sam598 but it didn't work for me. 

    You are right with your hint to the clipping problem and the switching framerate in low lighting condition. I considered it in that little experiment.

    Maybe a few words to the experiment itself.

    i pointed the kinect to a bright wall 2 meters away. i also put a white paper in the field of view of the kinect (approx 0.2m distance) as you can see in image1 on the left side of the image.

    After a short time (approx 50 frames) i moved a dark object in front of the wall and behind the paper as you can see in image2 . without exposure correction the luminosity of the white paper should stay constant. as you can see in the image of my first posting, the luminosity is nearly constant until the object enters the scene. The 2 curves of luminosity and exposuretime seems really similar thats why i asume a possibility to find a mathematical function.

    I continued the experiment several times under different lighting conditions. The results are similar to the image in my first posting which means that both graphs (luminosity and exposuretime) seems related.

    fyi: luminosity means the intensity of the L-channel from Lab-transformed image.

    image1

    image2

    Tuesday, September 15, 2015 1:50 PM
  • Thanks Sascha for your detailed experiments.

    It seems that, they uses a measurement based on the intensity (luminance) of captured pixels and try to set relative exposure time to have a total intensity fix or in a predefined range.

    Have you found any further facts? for example the most important part of the image?
    I mean I want to find a tricky way for setting up my environment which makes the Color camera works in my desired range of exposure time.

    Thanks in advance


    • Edited by Moha Es Saturday, June 17, 2017 5:06 PM
    Saturday, June 17, 2017 5:05 PM