View my account

Oscillations in the distribution of depth

Follow
Renaud Bastien

Hi,

we noticed an unexpected problem while trying to track a fish in a bowl. When looking at the distribution of depth values, we can observe oscillations where some values are more often found than others.

This is unexpected because our fish is supposed to explore the whole space and not some discreet positions in the vertical direction.

To confirm that the problem was coming from the camera, we designed a simple protocol. We attached a small red LED on a CNC rail, that is moving away from the camera. The LED is moving 1mm per step and at each step we are acquiring 10000 images.

This simple protocol gives rise to similar oscillations. We are a bit lost here. We haven't seen anybody reporting similar issues. Would there be a simple explanation why the depth distribution is not homogeneous?

thanks

1

Comments

3 comments

Date Votes
  • MartyG

    Hi Renaud Bastien  The camera can have difficulty reading depth detail from reflective surfaces such as the glass of fish aquariums.  If the camera is outside of the bowl then aligning depth to color can help to improve results.

     

    If the camera is immersed in water then the camera may need to be calibrated for that environment in order to account for distortion caused by the water.  There is an extensive technical discussion about this subject at the link below.

    https://github.com/IntelRealSense/realsense-ros/issues/1723

     

    The project FishSense is an example of fish monitoring and measuring with an immersed RealSense camera.

    https://www.youtube.com/watch?v=KA_x03_Aru8

     

    Which RealSense camera model are you using, please?

    0
    Comment actions Permalink
  • Renaud Bastien

    I am using a D435. The problem is not only seen with the fish but also with a single LED outside of water. We only use the stereoscopy and only the LED can be seen on the image. So there is only one small bright spot visible

    0
    Comment actions Permalink
  • MartyG

    There was a recent case at the link below where an LED was being measured at 60 cm distance from the camera and they were experiencing "depth jitter" variation in the measurements.  Their depth image showed that the bright LED light showed up but the rest of the scene's depth detail was mostly missing.

    https://github.com/IntelRealSense/librealsense/issues/10429#issuecomment-1107853739

     

    They confirmed that depth measurement jitter was significantly reduced if the camera's IR dot pattern projection was enabled.  Is the projector on your own camera enabled or disabled, please?

    0
    Comment actions Permalink

Please sign in to leave a comment.

Didn't find what you were looking for?

New post