I am looking into the possibility of photon for the next season but we will not be able to get an orange pi or raspberry pi. Do we need these things or are there cheaper alternatives?
According to the docs a raspberry pi 3 is the minimum recommended coprocessor.
Outside of that you also need an sd card and a camera (for april tags, ideally global shutter but not strictly required)
Would you know of anything that we could use that’s similar to photon?
I don’t think there are cheaper alternatives unfortunately. Photonvision is basically just the cost of the hardware since all the software work is done by volunteers.
Any other solution will probably be more expensive because you will be paying for the software as well (e.g. limelight)
Okay thank you
What functions do you want similar to PV? You can use a roboRIO for some of the functions and a PC for some other functions but a few functions may still be missing.
My team might have some RPi 3 somethings to dispose of. I can ask the coach if you are seriously interested. Also, is your organization a non-profit just in case coach says we can dispose of them at less than their value (looks like $35 to $45 plus shipping)? What’s your schedule? Not much happening this week with the holiday.
One other thing you would need is a power supply for the coprocessor. Something like this would work for a Raspberry Pi 3B, or even this at lower cost, but less brownout headroom.
I’m interested in why you could not get a Raspberry Pi or Orange Pi for a vision solution. If you went with a Raspberry Pi, an Arducam OV9281, a microSD card, and a power supply, you are probably in less than $120. The software is free. One Kraken motor is $200. A NEO/SparkMAX is $138. A 47" stick of 2x1 punched aluminum tube is $40. I know budgets can be tight and inflation has hurt teams, but when prioritizing the entire team spend, it seems like a vision solution should work its way higher on the list.
I strongly recommend sticking with the hardware on this page:
The Orange Pi 5 will last you forever. You can get away with running PV on a Raspberry Pi and any USB camera for detection, but you’ll have an easier time if your hardware is the stuff recommended by PV docs. Kludging something together is good for a first pass at trying it out, but you’ll want the good hardware for your final implementation.
Driver camera can be a Lifecam though. You definitely don’t need an OV9xxx camera for that.
The main purpose would be to read April tags for auto can that be done through the rio and thanks for the offer I will definitely talk to my people about it
Item | Cost | Notes | Link |
---|---|---|---|
Orange Pi 5 | $81.99 | Each Pi can support 2 cameras | Amazon.com |
Redux Zinc Power Supply | $20 | Each Zinc supports 1 pi | Zinc-V 5V/4A Step-Down Regulator |
Camera x2 | $49.99 | Black Friday Deal 42.49 | https://www.amazon.com/Arducam-Distortion-Microphones-Computer-Raspberry/dp/B096M5DKY6/ |
NVME SSD | $20-100 | MicroSD card can (and for us, did) pop out during competition. SSD is more secure | https://www.amazon.com/SN740-256GB-2230-30mm-NVMe/dp/B0C6MVP42M |
Total | $172 - $300 |
For anyone else coming across this thread who can get access to an orange pi, here’s a basic setup shopping list
There is a WPILib example vision program for reading AprilTags. I have a github directory that is based on that tag detection. I added code to calculate the angle the tag is from the center of the camera (principle point) - that’s only a few lines of code and requires a reasonably close estimate of the focal length. Also, I added the 3-D pose estimation - that’s a lot more lines of code.
The detection and 3-D pose estimation uses almost all of the CPU time of a roboRIO v1. If you have a roboRIO v2 and limit yourself to just “turn toward the tag” then it probably will run okay on a roboRIO v2.
You could assume that the 2025 game will have AprilTags not too far from the game piece targets and a small offset from the camera center will line you up.
Thank you for this
I have a student working on a 3-D printed case for this.
Please post a link here when they are done! Will it be a generic mount or something specific to your frame/robot?
As an aside: If you’re budget strapped and trying to maximize flexibility on your purchases, grabbing a color version of the Arducam might be a good idea.
https://www.amazon.com/dp/B0CLXZ29F9
This has all the same benefits of the standard Arducam (high fps, high fov, and global shutter), but at $60 vs $50, adds color. This makes it viable for driver camera, basic color pipelines, or ML object identification models. Still works equally well for pointing at april tags.
It’s not on the official support list, but works great for all these in our testing.
Can confirm, the 9782 rocks. Looks like it’s come down in price too, it used to be $90-100.
Trying to start with a simple box for handling and strain relief to start the rookie team on vision.
Don’t forget this clever device and a search of the Internet can turn up other working designs a lot quicker than waiting for us.