Getting started

Minimal requirements:

  • Raspberry Pi with Raspberry Pi OS 64 bit installed (both desktop and lite version are fine). We have tested it with kernel 6.1.21-v8+.
  • Raspberry Pi camera (e.g. v2, v3, HQ or global shutter) attached and working with libcamera.

On the Raspberry Pi

Start by SSH'ing into your Raspberry Pi.

Then on the Raspberry Pi

sudo apt install git cmake libgpiod-dev libcamera-dev

and clone our code:

git clone https://github.com/MayFly-AI/mayfly.git

Configure with CMake:

cd mayfly;
mkdir build; cd build;
cmake ..

Then compile:

make mayfly

Now start the executable and provide the path to the simple_camera example config.json file:

./bin/mayfly ../python/examples/simple_camera/config.json

This config.json configures a server to stream a 640x480 @ 30 FPS camera stream on port 8999.

On your machine

Download the same code to the machine you wish to stream data to:

git clone https://github.com/MayFly-AI/mayfly.git

Linux

On linux, from mayfly folder:

pip install .

Now enter examples folder:

cd python/examples/simple_camera

Find the IP address of your Raspberry Pi (you can use ifconfig) and run the simple_camera example providing this IP:

python simple_camera.py --ip XXX.XXX.XXX.XXX --port 8999

A cv2 windows should appear with the live stream.

The example simple_camera.py uses the OpenH264 decoder. If you have a NVIDIA GPU, we recommend using the nvdecode decoder. To do this, run:

python simple_camera_cuda.py --ip XXX.XXX.XXX.XXX --port 8999

Recommendations

  • We recommend the CM4 with external WiFi antenna for good WiFi performance.

  • We recommend using a dedicated access point. We have good experience with ASUS ZenWiFi Pro ET12. On the access point, we disable the 2.4GHz and 6Ghz bands and for the 5Ghz band, we set it to 80MHz bandwith and find an undisturbed channel. On the RPI4, check the established wifi connection with

    iw dev wlan0 info

    It should list the channel and 80MHz bandwith.