View my account

D435 Area without any depth information

Comments

11 comments

  • MartyG

    Hi Olivier Huet   I have examined your images very carefully.  Would it be possible please to provide more detail about how your program works in order to help me to understand what it represents.  It looks as though it is prioritising the ground detail close to the camera (which is what I would expect, as you stated that it is a ground detection application) and giving lesser priority to the non-ground detail.

    It does not look as though alignment, such as depth-to-color, is being applied to the RGB and depth images.

    If this was a normal depth image, I would speculate that the missing detail on the left side was due to a phenomenon called the invalid depth band, where the left end of a depth image can appear rough-edged.  The invalid depth band is described on pages 68-69 of the current edition of the data sheet document for the 400 Series cameras, and an image from that section of the document is shown below.

    https://dev.intelrealsense.com/docs/intel-realsense-d400-series-product-family-datasheet

     

    0
    Comment actions Permalink
  • Olivier Huet

    Thank you for your reply. I send you the raw depth files as recorded by the camera.  

    https://1drv.ms/u/s!AvCN2MMBG_5EuhqEwyXql9MOimE0?e=CRoOop 

    The phantom area appears as a blue area on the depth image colorized by opencv. Yes it is an invalid depth zone but not a band at all. 

    It's more like a shadow or a cloud. The ground is free at this location. Like you can see on the RGB picture 

    I have thousands of images like this ... with a ghost on the depth image. 

    On this picture i have paint the area in black 

    https://1drv.ms/u/s!AvCN2MMBG_5EuhhF6EufcbFe3nK1?e=TqtNel 

    0
    Comment actions Permalink
  • MartyG

    Thanks very much for the files.  Is the grid of squares applied as an overlay onto the images after they have been generated?  In other words, it is not something that the camera sees.

    The empty space reminds me of an effect on depth images that I have seen in a couple of recent cases where an area near to the camera on the depth image was empty or excessively noisy.  In those cases, changing the depth unit scale of the camera from its default of 0.001 to a smaller value of 0.0001 greatly improved the depth image in the affected near-range area of the image.  It has the effect of making the image's viewpoint slightly more zoomed-in, but this may not matter if you are scanning the ground in the center of the viewpoint and at short range.

    0
    Comment actions Permalink
  • Olivier Huet

    The squares of the grid are used for debugging, the camera does not see it, it is indeed an overlay.

    I going to try your solution.

     

    0
    Comment actions Permalink
  • Olivier Huet

    Hello MartyG

    I am a little confused. changing the depth unit with Android software does not seem possible. I tried to patch the JNI C code but the result is not really relevant. Can you please tell me the correct way to change the depth unit scale when using android api?

    Olivier

     

    0
    Comment actions Permalink
  • MartyG

    Hi Olivier Huet  JNI supports the loading of custom Visual Preset (json) camera configuration files.

    https://github.com/IntelRealSense/librealsense/pull/3771

    I believe that that the json parameter that would set the Advanced Mode depth unit scale would be param-depthunits

     

     

    You can create  a custom json file though without hand-programming it by defining camera settings in the RealSense Viewer program and then using the json export button near the top of the Viewer's options side-panel to export the settings as a json that can then be loaded into your Android application with scripting.

    The Depth Units setting can be found under Stereo Module > Advanced Controls > Depth Table

     

     

     

    The default value of param-depthunits is '1000'.  Changing it to '100' creates a zoomed view similar to changing the depth unit scale option to 0.0001 earlier in this discussion.

     

    0
    Comment actions Permalink
  • Gaetan Pardon

    hello

    The problem is getting worse.

    The IR camera is saturated with daylight.

    With summer approaching, it's only getting worse.

    this problem does not appear on the D415.

    it only appears with the D435.

    we are using the latest firmware.

    Our tests were carried out under the same conditions for the two cameras.

    They were carried out on two supports on smartphone(the same as before) and on computer(Windows 10).

    They were repeated with different versions of the firware.

    In all of these cases, only the D435 encountered this problem.

    The sample image was taken at 9 a.m.

    do you have a solution for this problem ?

    gaetan interns at Perception

    0
    Comment actions Permalink
  • MartyG

    The D415 vs D435 issue may be related to the D435 having wider infrared sensors than those on the D415.

    If auto-exposure is enabled then you could next try defining a Region of Interest (ROI) to alter how the camera deals with outdoor sunlight, as described in Point 3 of Intel's tuning-guide information on using a camera in sunlight.

    https://dev.intelrealsense.com/docs/tuning-depth-cameras-for-best-performance#section-use-sunlight-but-avoid-glare

    0
    Comment actions Permalink
  • Gaetan Pardon

    Hello

    We have already deactivated auto exposure and set gain and exposure to minimum without that being enough.

    Do you have another idea ?

    gaetan interns at Perception

    0
    Comment actions Permalink
  • MartyG

    I performed further tests with D415 and D435 whilst pointing the camera at a bright light source.  Enabling the camera's HDR mode produced lines on the D435 depth image similar to the ones in your image above, whilst on D415 it significantly improved the depth image.

    Is HDR mode enabled in your application, please?

     

    D415 (UPPER) AND D435 (LOWER)

     

    0
    Comment actions Permalink
  • Gaetan Pardon
     
    Hello
     
    Currently we do not use HDR. So it can not be the cause of our problem.
    Do you have another idea?
     
    gaetan interns at Perception
    0
    Comment actions Permalink

Please sign in to leave a comment.