Raspberry Pi 4 as coprocessor

Hello teams, hope you are doing great!
So, our team wants to do vision this year, and we were wondering if it is a viable option to connect a raspberry pi 4 as a coprocessor for our roborio in order to analyze the video streams of the robot and publish values to network tables using programs like GRIP, and if there are any better or cheaper options (all the r-pi 4’s i’ve found in amazon are around 75 bucks), thanks in advance!

The raspberry pi 4 sells for $35 from multiple vendors; for example Adafruit. There’s a list on the Raspberry Pi website.

To add to the questions of OP, what’s the best way to power a Pi 4 on the bot; the power port is USB-C.

1 Like

https://www.amazon.com/Raspberry-Model-2019-Quad-Bluetooth/dp/B07TD43PDZ

$40 at Amazon. Might be cheaper elsewhere.

1 Like

While the Rpi 4 is a viable option, it does not contain a gpu or much processing power for vision. Instead, I would recommend getting the Jetson Nano, which is around $90. It has a 124 CUDA gpu and is made for vision processing and AI development. If you are doing very minimal vision the Rpi is fine, anything bigger you will want a Nano to get a decent FPS.
(Edit)
To add on, the nano can also be powered using PoE (Power over Ethernet) directly from the RoboRIO

1 Like

The Raspberry Pi 4 has more than enough power for vision processing. The Limelight runs on a Raspberry Pi (and capable of 90 fps tracking). Other solutions such as Chameleon Vision also run on this hardware and we have achieved processing at 187 fps.

3 Likes

The other advantage of Raspberry Pi is that the WPILib team provides an image for vision processing.

2 Likes

thank you so much!!!

quick question tho, is 1GB, 2GB or 4GB the best option, cause prices change a lot :S https://www.amazon.com/Raspberry-Model-2019-Quad-Bluetooth/dp/B07TC2BK1X?th=1

We just bought the cheapest one (1 GB) for testing a couple months ago and that’s been working well for us.

1 Like

okok tysm for the info!!!

Here is the official documentation for doing vision on a Pi: https://docs.wpilib.org/en/latest/docs/software/vision-processing/raspberry-pi/index.html

1 Like

After getting the WPI Image on the Raspberry PI, I wrote a page to take one of the examples and getting OpenCV working on it.

Extending the WPI cpp-example on the Raspberry PI

1 Like

May I ask how you managed this? When our team tried a rasberry pi we got less than 30 FPS

1 Like

Check out Chameleon Vision: Chameleon vision 2.0

2 Likes

I may also interject with OpenSight here, it allows for more customizable and complex pipelines like GRIP compared to other vision suites.

1 Like

Huh, I didn’t know any programs existed for helping with Vision. My team has always used Tensorflow to make our own, but it gets complicated fast. Thank you, we might just check this out :grin:

1 Like

One idea is to decrease the resolution of the image before processing. The limelight processes 320 x 240 images, while the lifecam is capable of 1280 x 720.

2 Likes

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