Machine Learning/Computer Vision for Cargo Recognition

There will be a new Pi image hopefully this weekend, I’m waiting for RobotPy’s 2022 initial release.


Yeah, just as a follow-up to this, just ran a pi4b with 8gbs of ram and no coral and got on average ~3.5fps on the Axon tracking at 480p lol.

No coral is painful, especially right now

thanks for doing that… kinda confirmed my fear. Looks like computer vision it is with corals not able to be bought (or if they are for a ridiculous pross).

I can try it with my Coral or Jetson Nano if you could get me the model. Otherwise I’ll try to build one myself.

I saw khadas has a Am311D NPU board for $99 on Amazon. It’d be custom code again, but perhaps worth a shot. Khadas VIM3 Development Learning Board NPU Recognition Amlogic A311D Amlogic TV Box Set-Top Boxes : Electronics

1 Like

I tested Photonvision’s colored shape tonight and it worked wonderfully for detecting the balls. If you are doing this just to learn cool new things I am completely on board and wish you the best of luck. If you think this will give you a competitive advantage I fail to see how that is the case.

I do vision ML in my day job, so I have significant experience with ML. Uniformly colored spheres on a solid background is a solved problem with machine vision and IMO Photonvision is the quickest way to implement that.


Even raspberry pi 3s are hard to find and PhotonVision is not accelerated on anything but that.

1 Like

Are you saying you have tested it and the framerate you are getting is insufficient for tracking balls (I find this hard to believe)? If not what are you saying?

1 Like

It looks like the field has a fence of poly-carbonate.

Did you test if it can determine if its a reflected ball or real ball?

Did you test where one ball partially obscures another ball?

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.