Hi

  1. We are using PoE models. And We frequently meet an issue which is IP conflicts during the PoE camera connection at the initial stage. Do you have any solution to avoid this problem?

  2. The depth of OAK-D, which can be obtained through DepthAI, is output in discrete form of a specific level, but we are wondering whether it can not be obtained in a continuous form as much as possible or not??? Or such as do you have any API to set up this?? (For example, in our case that needs to cover a curved surface as well)

  3. Is there an option that can be utilized when 3D Projection work is required?? We are using Open3D by default, but there may be other alternatives??

  4. Could you share the inforamtion of the API to access camera settings such as luminance and contrast with code??

  5. We hope to have the function with the higher the dynamic range. Could you have any API or solution or advice to enhance its performances?

  6. We hope to have the camera lens less distortion. Do you have any API or solution or advice for this issue?

  7. We hope to have the less noise with depth function. Do you have any API or solution or advice for this issue?

Best regards,
Ryan.

  • erik replied to this.

    Hi RyanLee ,

    1. What IP issues do you mean? If it's general connection instability with POE, we have just pushed xlink fixes that should help with instability. These will get merged soon.
    2. You can enable subpixel mode, which adds additional subpixel bits (up to 5 bits) to disparity, which results in a much more "continuous depth".
    3. You could also use eg. PCL library which is quite popular in robotics.
    4. You can change ISO sensitivity and exposure time, API docs here.
    5. RVC2 doesn't support HDR, but RVC3 will support it.
    6. Is that distortion of the wide FOV lenses? If so, you can check demo here.
    7. This really depends on the environment. If you have good lighting/texture, you can look into all the bits and pieces of depth configuration and postprocessing filters.

    Thoughts?
    Thanks, Erik

        erik

        I have one more question.
        which is the better performance between PCL and Open3D?? Or which one is more used in robotics??

        Best regards,

        • erik replied to this.

            Hi RyanLee ,
            I think it really depends on the function you use - they are both implemented in C++ (open3d has python bindings), and afaik PCL is more used in combination with ROS.
            Thanks, Erik

              Hi Erik,

              "You can enable subpixel mode, which adds additional subpixel bits (up to 5 bits) to disparity, which results in a much more "continuous depth"."

              We have some more question about it. Because we alredy test it. So if it is possible, we hope to have some more advice to enhance its depth perfomance like continuous depth as much as we can. Do you have any idea about it??

              Best regards,
              Ryan.

              • erik replied to this.

                RyanLee Have you tried 5 subpixels bits already? As it's 3 subpixels by default. I don't know any other trick to have a more "conitnous" depth.
                Thanks, Erik