Has anybody tried using the Orange Pi 5+ for vision? We’re having a bit of trouble finding an acceptable OS image to use (especially one that doesn’t use Chinese repositories). And, any opinion on cameras?
2 Likes
Photonvision says they support Orange Pi 5. You might need to read into the documentation to know specifically if this answers your question.
There are links to a variety of OS images at https://docs.photonvision.org/en/latest/docs/getting-started/installation/sw_install/orange-pi.html
We are experimenting with the Debian image. Arducam cameras are working very well for us. We are using black and white as our goal is to read April Tags so colour isn’t an issue. We are hoping to release a white paper in a few months but there is lots of work to do!
You can find my guide here: Orange Pi Guide For Photonvision 2023 - Google Docs
I highly recommend the Orange Pi 5 for FRC.
The parent folder has more info on cameras: Coprocessor - Google Drive
8 Likes
For this year’s competition season, we ran two Arducam OV9281 global shutter cameras with PV for AprilTag tracking on a RPi4 powered off an Anker battery pack providing USB-C at 3A.
For our off-season events we switched over to the Orange Pi 5 so we could run the PV cameras at higher resolution for better range while switching over to a dual output cable from two ports on the Anker battery pack (providing the 4A that the Orange Pi 5 theoretically could use with various USB ports in use).
We recently started having PV configuration corruption/reset issues where after a robot hard power cycle, one or both of the camera configs would be lost and need to be restored by hand (calibration import, switching to AprilTag pipelines, etc.). We weren’t sure if this issue started happening because of the switch to Orange Pi from RPi or our Anker batteries getting “tired” from constant drain/recharge cycles over the season and causing power drops to the OPi.
At our last off-season event this weekend, we switched over to running the OPi off one of the low power constant current channels on the REV power distribution hub using a 12V - 5V / 5A buck converter to at least take the battery power source out of the troubleshooting equation. We experienced the same issue where at least one camera would routinely lose its configuration (once or twice both cameras were lost). We then just hand configured the cameras via PV right before each match and left the robot powered on.
When the cameras were configured correctly and power not cycled, our vision worked fine.
Reading more today about other teams experience and notes with PV on mini PCs (Beelink, etc.), it looks like we may need to explore making the Orange Pi / Debian OS go into readonly mode (using overlayroot) after PV and camera configuration is dialed in so that each power cycle/reboot operation comes back with essentially a “factory” file configuration. This is next up for us during off season research/testing.
In short, the Orange Pi + PV setup has been working perfectly for our use of AprilTag tracking as long as our camera/pipeline/etc. configurations are intact after boot. (We are using AprilTag tracking for both automated scoring sequences - identifying if the robot is in front of a cone or cube node - and for fine-grained auto alignment of the robot to the nearest node on all three axes).
Hope this helps.
I believe that the settings loss issue should be fixed this season, at least. Read only mode can be a bit of pain when you actually need to change exposure or something.
1 Like
We hope this issue is fixed in the 2024 release, pending the results of beta testing (coming soon).
I assume y’all are referring to: Robustly remember previous pipeline, resolution, etc. between system shutdowns · Issue #552 · PhotonVision/photonvision · GitHub
Yes, we haven’t been able to spend time yet really debugging/tracing whether our issue was truly a PV one or something in our OrangePi / power / etc. setup. This older bug entry seems to indicate it could likely be PV.
nah this PR Add SQL config manager by mcm001 · Pull Request #818 · PhotonVision/photonvision · GitHub
It appears to have started this (2023) season, and the best explanation we have is a quiet change in Debian or another part of Linux everyone uses that makes the FS less robust under FRC loads
2 Likes
Would you use the Orange Pi 5, Pi 5 Plus, or the Pi 5B? Seems the 5 Plus and 5B are more available than the regular 5.
These are not images for the plus variant. The links to the various OSs can be found here, and here’s a direct link to the Ubuntu images. Make sure to select the actual version of Ubuntu and Server/Desktop you need. For PV you should get Ubuntu 22.02 server.
There is this project for non official ubuntu images which is nice. Personally I had a feature missing there called RGA but it’s an edge use case.
As long as it uses an RK3588S or RK3588 it’ll work the same. I use a 5 (not B or Plus) with 4GB RAM because it’s the cheapest and I don’t need WiFi or Bluetooth. I would recommend the same so that you don’t accidentally enable the wireless modes, but it can make setup easier if you don’t have a Wifi extender or something with an Ethernet port to connect to for setup.
Both the 5B and 5 Plus use the RK3588 or RK3588S. We are looking at getting one with 16GB of RAM. It’s a little more expensive, but hopefully, it will give us a little longer use. The onboard wi-fi was giving me pause, but it seems both the 5B and plus have it built in. The 5 seems to be harder to get than either of the others currently. Do you have an opinion between the 5B and 5 Plus?
What are you planning to run on it that you think you’ll need more than. 8Gb of RAM?
Gotta be able to stream Family Guy Clips and Subway Surfer Playthrough to keep the students interested.
Just planning to run Photonvision. We likely don’t need more than 8GB, but it’s not much more to get 16. The hope there is that it would keep us from having to buy yet another vision processor next year.
The current cv2 vision algorithms people are using won’t ‘scale’ with more ram. 8GB is probably overkill as it is if all your doing is running an apriltag algorithm. The processor speed is going to be your bottle neck.
1 Like
Yeah I ran a Pi 4 1G and had no issues, I don’t even think it passed 60% utilization when we were testing
I’d believe it! haha the reality is your processing fairly small images, its just more about the math and computation being done on them than the memory storage
I’d save some money on the ram size and put it away to upgrade the processor in a few years
I’ve been using that project and, so far as I can tell, it works well. Has the advantage that it’s not using the Huawei repositories.
The 5 Plus does not have wifi built-in – it has an m.2 e-key that allows you to connect a wifi/Bluetooth daughter card. But, if you’re just going to stick it in on a rio, better to just not use that.
The 5 Plus uses the RK3588 v. the RK3588S that’s on the 5B and original 5. In theory, that gives the 3588 faster video over MIPI CSI, but it seems that most people are using the board with a USB Camera, so that’s not really a great advantage.
At this point, I think we’llLimelight, an integrated vision coprocessor probably go with the Raspberry 5, even though it has fewer processor cores, mainly because the ecosystem is far more robust. (Digikey has a bunch of the 4GB models in stock).