Hi there,

I'm currently using an OAK1 with a Raspberry PI 4 (2Gb) for initial tests and wanted to upgrade to a better system for object inspection. A first idea was to have a system as shown below:

  • Two OAK1 cameras in parallel, so I'd have approximately 2px/mm with a 4K resolution
  • A Raspberry Pi 4 (8Gb)

I don't have the need of a high fps inference (something like 0.3fps should work fine).
Any ideas/limitations on the construction? Using some Jetson board instead of a Raspberry could be helpful?

Thanks in advance!

  • erik replied to this.

    Hello jvitordm ,
    I think even RPi could be able to handle results if you run your AI models on the OAK-1 camera directly. What kind of AI model would you like to use? For the construction, it might be that 2m USB3 cable (5Gbps) will have issues with EMC, so you might need to use USB2 mode (480Mbps).
    Thoughts?
    Thanks, Erik

        erik

        Perfect, I think I'll try it with the RPi to start. The idea is to use a Mask RCNN instance segmentation model - do you think it's possible to run it with such a great resolution (4K) if I have a low demand on fps (lower than 1fps)?

        About the cable, 1m should do if a put the RPi centered and the cameras 0,95m away, so that shouldn't be a problem.

        Thanks once again!

            Hello jvitordm ,
            I'm quite sure 4k won't be able to run even below 1FPS. At 300x300 it runs at about 2.2FPS, see demo here. I would suggest waiting for the KeemBay to be able to run such model on 4k images.
            Thanks, Erik