• DepthAI-v2

  • any OAK Series for smart farm application

We are currently reviewing the implementation of a smart farm with the oak series. Could you please share any information or experiences you have listed? If there any any similar example or experience, it will be very helpful for us to go the next step definitely.

<Questions>

  1. Distance measurement
  2. Measure the size of a moving object
  3. Maximum measurable distance
  4. Maximum measurable distance
  5. Minimum measurable distance
  6. Feed measurement in place in the dark environment of the farm

And i saw some of your sample application, but i hope to have the best approach to work efficiently with your help.

Thanks.

  • erik replied to this.

    Hello RyanLee, could you elaborate on the application, what would the OAK camera do in this smart farm (so we can share information/experience on this)?

    1. If it's an object (eg. apple) and you will use Yolo or MobileNet NN, you can use YoloSpatialDetectionNetwork or MobileNetSpatialDetectionNetwork. This will detect the object and return XYZ coordinates of that object, and you can use Z for distance measurement. Otherwise, you could use SpatialLocationCalculator and select ROI of the area of which distance you are interested in.
    2. What would the object be like? A box, with borders or eg. an apple? For example, here's an approach someone took to measure box size with OAK cameras. But it might have to be a quite different approach if it weren't a box.
      3./4. All our devices have max depth perception of 35m, except OAK-D-Lite, which has 18m.
    3. About 20cm with extended disparity mode and at 400P. More info here.
    4. Passive stereo won't work in a dark environment, and for that reason, we are designing OAK-D-Pro / OAK-D-Pro-POE which will have an on-board IR laser dor projector (better depth quality) and IR illumination LED (for working in dark environments).

    Thanks, Erik

      5 days later

      Hi, Erik. Thank you for your response quicky and correctly.
      And we are using it for measuring the size of chickens at the farm and checking the remaining feed amount in the large round barrel.
      If you have any experiences with this, please share with us to have better idea for our project.

      And we made some questions for our project to check its compatibility with OAK-D or OAK-D-Lite series.

      1. Could you share IP level of OAK-D-Lite
      2. And we need Wi-Fi and Bluetooth to communicate with our server. Could you suggest a hardware for it? And I know one device OAK-D-IOT 75. Could you share its price and when it is available to buy for us??
      3. We are using detectron2 with PyTorch. Is it possible to convert it for OAK-D-Lite??
      4. Could you share the power consumption of OAK-D and OAK-D-Lite during the deeplearning model is working?
      5. Is OAK-D or DAK-D-Lite have a sleep mode? If so, how can we control it?? And during the sleep mode how much current is consuming?
      6. The front of the camera is covered with dust due to dust in the farm. Have you had any experience in solving this problem?

      Thanks,

      • erik replied to this.

        Hello RyanLee, happy to help! We don't have any experience measuring chicken sizes, do you mean heigth/width? We plan on creating such demos soon, but initially for cuboids. The remaining feed amount would be much easier, and I assume depth wouldn't be needed (object detection + remaining feed recognition model).

        1. We haven't tested it yet, but we didn't design the enclosure to be waterproof/dustproof
        2. OAK-D-IOT-75 is $199 (shop). Shipping will be in December. You could also use eg. OAK-D-Lite with RPI 4b, which has both wifi and Bluetooth connectivity
        3. I would assume it would be possible (possibly with some editing of the model in case some layers aren't supported) but that it is too heavy for our current devices. I couldn't find any FPS info, could you share that? Also, our future devices will be based on Intel's Keembay VPU (next gen of Myriad X), which we assume will be 10x faster for inference, and could run such heavy models. ETA Q2 of 2022.
        4. about 4W to 4.5W if you are running video encoding, stereo depth, image manip etc. as well.
        5. Not yet, but if you don't run anything (device in standby) the power consumption is about 0.5W
        6. We haven't, but this might be problematic for the stereo depth perception. What was the current solution to such problem?

        Thanks, Erik

          6 days later

          Hi

          Thank you for your valuable information.
          Here is answer according to your feedback in the previous.

          We don't have any experience measuring chicken sizes, do you mean heigth/width?
          ==> Yes, we mean the height and width.

          We haven't, but this might be problematic for the stereo depth perception. What was the current solution to such problem?
          ==> we also don't have any solution for it so we are researching at the same time asked you to get any experience if you have.

          We plan on creating such demos soon,
          ==> When it is ready, please share with us. That would be great for us.

          Additional question.
          Could we have the size information about OAK-D-PRO-POE? is it the same with OAK-D-PRO from https://github.com/luxonis/depthai-hardware/issues/114 ??

          Thanks.

          • erik replied to this.

            Hello RyanLee , sounds great!
            1) Chicken sizes - to add to my comment above, if someone does try to achieve work on this, an interesting approach would be rotating calipers method. We will experiment with this approach soon.
            2) Dust - I would first try it out, as it might not be such a problem, but really depends on the dust size/volume. I would imagine stereo depth perception will become more noisy, where proper filtering would help.
            3) We will - just added a note on this task.
            4) I think it will be a bit bigger, but still much smaller compared to the original OAK-d-POE. I will ask the team for step files.
            Thanks, Erik