Is it possible to do machine learning so that one can recognize a note? we personally have a pi 5 module connected to a usb cam that runs photonvision and i was wondering if recognizing notes were possible. maybe via a custom pipeline? i am very much a beginner on vision processing.
Check out this thread Tracking Notes with Vision (Demo, Model, Tools)
And then go to this
There’s a few others as well if you use the search at the top right.
Best of luck!
3 Likes
This only runs on an orange pi, not rpi 5, idk what pi 5 op has. Otherwise you could run a color detection pipeline for the color orange, it’s easier and would work fine as long as you tune it to not see red bumpers.
Yup my bad, I misread OP referring to an orange pi. Kinda forgot an rpi5 is now out.
1 Like
The pi5 should be able to run TFLite models at a respectable speed - Raspberry Pi - Ultralytics YOLOv8 Docs should walk through how to do it.
2 Likes
Alternatively if you’re willing to spend a little more on top of your raspberry pi’s, you can also connect a Google Coral TPU to your pi, either with the Coral USB acclerator or the M.2 slot on the RPi5, and run a TFLite model on a dedicated TPU.
Hey guys, I have noticed a lot of teams using ML for note detection. Why not just use colored shape detection from photonvision? Colored Shape Detection - PhotonVision Docs. It seems to be much simpler and doesn’t require a raspberry pi.
The notes are a nice orange, so colour detection is possible and we already have it working and the robot tracking the note.
The issue comes from the orange also looking the same as red and there is quite a bit of red on the field from the tape to bumpers which get falsely detected. You can work hard and spend a lot of time tuning the colour range which may or may not clean it up enough. Spending the effort of getting some great values that work at home may be null and void because the lighting conditions at comp is different.
Machine learning just adds more parameters to detecting the note than just normal colour detection.
Instead of just looking at is it between these colour ranges, we look at the shape, the size, the colour, the colour gradient and whatever other parameters the ai picks up on to identify a note.
1 Like
Please note that this fork is unofficial and users won’t be technically supported from the PV team (we don’t know if the code has deviated, what changes have been made, etc.)
PhotonVision also now supports orange pi based model detection as well.
3 Likes
This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.