View my account

InterRelasense D455 depth quality

Comments

7 comments

  • MartyG

    Hi Maiskhtb  If a physical filter product called a thin-film linear polarization filter is applied to the outside of the camera, overlaid over the sensors on the front of the camera, then it can significantly dampen glare from surface reflections and so improve image quality.  Most polarizing filters will work as long as they are linear, so they are inexpensive to purchase.  You can find them on stores such as Amazon by searching for the term linear polarizing filter sheet

     

    Also, does depth stability improve and holes reduce if you enable some post-processing filters in your ROS launch?

    ROS1

    roslaunch realsense2_camera rs_camera.launch filters:=temporal,spatial,hole_filling

    ROS2

    ros2 launch realsense2_camera rs_launch.py hole_filling_filter.enable:=true temporal_filter.enable:=true spatial_filter.enable:=true
    0
    Comment actions Permalink
  • Maiskhtb

    thank you so much MartyG for your suggestions! I will buy them and try!

    As for the filters, I tried them in Realsense Viewer but it didn't improve the image holes that much.

    I also want to show the original RGB image corresponding to the above depth image, as you see, I have a basket that already had holes, so the holes in the depth image are correct, the ones that are not correct are the ones corresponding to the white oval object in the middle, how to improve the depth value? sorry for the bad quality. 

    0
    Comment actions Permalink
  • MartyG

    The areas around the center oval that show up as solid instead of holes seem to have something just below the holes on the underside of the grid on the RGB image that may be causing the camera to recognize those areas as a flat floor surface instead of a hole.  So the areas appear 'filled in' on the depth image.  This is typical for depth images of holes with shallow depth.

     

    Your situation reminds me of a case at the link below where there was a grid of bars in the real-world scene arranged in the horizontal and vertical directions, and the camera would often only detect the horizontal bars or the vertical ones but not both at the same time, depending on which way the camera was orientated.

    https://support.intelrealsense.com/hc/en-us/community/posts/29135135902611-Difference-between-horizontally-placed-steel-bar-and-verticall

     

    You could try tilting the camera at a diagonal angle (of no more than 30 degrees forward tilt) instead of facing it straight down towards the box to see if that improves the rendering of the grid pattern.

    0
    Comment actions Permalink
  • Maiskhtb

    well i tried to tile the camera, turn off the light, and add the filters you suggested, the depth is improved so much but there is strange shaking in the depth image, I uploaded in this link a video and an image for my situation, you can see a noticeable shaking on the bottom of the video

    I also added a video of when the light is turned on, I also wish to be able to improve and fix it as well:

    https://drive.google.com/drive/folders/1Kd3NVHlf751UScZmuR-ySvE6wPqBdhjW?usp=sharing

    0
    Comment actions Permalink
  • MartyG

    Looking at the area in the bottom where the shaking is worst and looking at the same area on the RGB mage, the dark gray area there might be contributing to the instability.  It is a general physics principle (not specific to RealSense) that dark gray and black colors absorb light, making it difficult for depth cameras to obtain detail from them.  Projecting a strong light onto such surfaces can help to bring out depth detail from them if it is required,

     

    Changing the Temporal filter's Filter Smooth Alpha setting from the default value of '0.4' to '0.1' can help to stabilize such flicker.  You can test this in the RealSense Viewer by expanding open the Temporal filter's controls in the Post Processing section of the side-panel.

    0
    Comment actions Permalink
  • Maiskhtb

    Thank you so much MartyG , I didn't think about it physically,

    Sorry for all the questions, but regarding the object on the right side of the image (inside the green hand-drawn circle), why is the depth so inaccurate? It's crucial for me to get the most accurate depth values and contours of the white objects, as I need to find their exact centers in real-world coordinates. What would you recommend to improve this? Would switching to another RealSense model, like the L515, help solve these issues?

    0
    Comment actions Permalink
  • MartyG

    The L515 is a retired camera model and is no longer supported in the RealSense SDK and RealSense Viewer. 

     

    The D455 that you currently have has a wide field of view size and your crate is only just fitting into its view, so another RealSense 400 Series model with a smaller view size may not be able to fit the entire crate into its viewpoint when the camera is placed at the same distance from the crate.

     

    The object at the side of the crate could be causing the camera's depth sensing some confusion as it is right beside the corner of the crate.  It is similar to how the corner of a room where two walls go off at right-angles will be harder for the camera to depth-analyze than facing the camera straight at one wall.  The circled inaccurate object is also positioned beside a corner.

     

    If it is not practical to move the object away from the corner of the crate then you could test in the RealSense Viewer whether using the Preset drop-down menu near the top of the options side-panel to select a different camera configuration preset improves the depth accuracy.  I recommend trying the Medium Density preset as it provides a good balance between accuracy and keeping a lot of detail on the depth image.

    0
    Comment actions Permalink

Please sign in to leave a comment.