AIRASPI Build Log

By Aron Petau • 6 minutes read •

AI-Raspi Build Log

This should document the rough steps to recreate airaspi as I go along.

Rough Idea: Build an edge device with image recognition and object detection capabilites.
It should be realtime, aiming for 30fps at 720p.
Portability and usage at installations is a priority, so it has to function without active internet connection and be as small as possible.
It would be a real Edge Device, with no computation happening in the cloud.

Inspo from: pose2art

Hardware

Setup

Most important sources used

coral.ai Jeff Geerling Frigate NVR

Raspberry Pi OS

I used the Raspberry Pi Imager to flash the latest Raspberry Pi OS Lite to a SD Card.

Needs to be Debian Bookworm.
Needs to be the full arm64 image (with desktop), otherwise you will get into camera driver hell. {: .notice}

Settings applied:

update

This is always good practice on a fresh install. It takes quite long with the full os image.

sudo apt update && sudo apt upgrade -y && sudo reboot

prep system for coral

Thanks again @Jeff Geerling, this is completely out of my comfort zone, I rely on people writing solid tutorials like this one.

# check kernel version
uname -a

# modify config.txt
sudo nano /boot/firmware/config.txt

While in the file, add the following lines:

kernel=kernel8.img
dtparam=pciex1
dtparam=pciex1_gen=2

Save and reboot:

sudo reboot

# check kernel version again
uname -a

edit /boot/firmware/cmdline.txt

sudo nano /boot/firmware/cmdline.txt

sudo reboot

change device tree

wrong device tree

The script simply did not work for me.

maybe this script is the issue? i will try again without it {: .notice}

curl https://gist.githubusercontent.com/dataslayermedia/714ec5a9601249d9ee754919dea49c7e/raw/32d21f73bd1ebb33854c2b059e94abe7767c3d7e/coral-ai-pcie-edge-tpu-raspberrypi-5-setup | sh

What to do instead?

Here, I followed Jeff Geerling down to the T. Please refer to his tutorial for more information.

In the meantime the Script got updated and it is now recommended again. {: .notice}

# Back up the current dtb
sudo cp /boot/firmware/bcm2712-rpi-5-b.dtb /boot/firmware/bcm2712-rpi-5-b.dtb.bak

# Decompile the current dtb (ignore warnings)
dtc -I dtb -O dts /boot/firmware/bcm2712-rpi-5-b.dtb -o ~/test.dts

# Edit the file
nano ~/test.dts

# Change the line: msi-parent = <0x2f>; (under `pcie@110000`)
# To: msi-parent = <0x66>;
# Then save the file.

# Recompile the dtb and move it back to the firmware directory
dtc -I dts -O dtb ~/test.dts -o ~/test.dtb
sudo mv ~/test.dtb /boot/firmware/bcm2712-rpi-5-b.dtb

Note: msi- parent sems to carry the value <0x2c> nowadays, cost me a few hours. {: .notice}

install apex driver

following instructions from coral.ai

echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list

curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -

sudo apt-get update

sudo apt-get install gasket-dkms libedgetpu1-std

sudo sh -c "echo 'SUBSYSTEM==\"apex\", MODE=\"0660\", GROUP=\"apex\"' >> /etc/udev/rules.d/65-apex.rules"

sudo groupadd apex

sudo adduser $USER apex

Verify with

lspci -nn | grep 089a

sudo reboot

confirm with, if the output is not /dev/apex_0, something went wrong

ls /dev/apex_0

Docker

Install docker, use the official instructions for debian.

curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh

# add user to docker group
sudo groupadd docker
sudo usermod -aG docker $USER

Probably a source with source .bashrc would be enough, but I rebooted anyways {: .notice}

sudo reboot

# verify with
docker run hello-world

set docker to start on boot

sudo systemctl enable docker.service
sudo systemctl enable containerd.service

Test the edge tpu

mkdir coraltest
cd coraltest
sudo nano Dockerfile

Into the new file, paste:

FROM debian:10

WORKDIR /home
ENV HOME /home
RUN cd ~
RUN apt-get update
RUN apt-get install -y git nano python3-pip python-dev pkg-config wget usbutils curl

RUN echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" \
| tee /etc/apt/sources.list.d/coral-edgetpu.list
RUN curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add -
RUN apt-get update
RUN apt-get install -y edgetpu-examples

# build the docker container
docker build -t "coral" .

# run the docker container
docker run -it --device /dev/apex_0:/dev/apex_0 coral /bin/bash

# run an inference example from within the container
python3 /usr/share/edgetpu/examples/classify_image.py --model /usr/share/edgetpu/examples/models/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite --label /usr/share/edgetpu/examples/models/inat_bird_labels.txt --image /usr/share/edgetpu/examples/images/bird.bmp

Here, you should see the inference results from the edge tpu with some confidence values.
If it ain't so, safest bet is a clean restart

Portainer

This is optional, gives you a browser gui for your various docker containers {: .notice}

Install portainer

docker volume create portainer_data
docker run -d -p 8000:8000 -p 9443:9443 --name portainer --restart=always -v /var/run/docker.sock:/var/run/docker.sock -v portainer_data:/data portainer/portainer-ce:latest

open portainer in browser and set admin password

vnc in raspi-config

optional, useful to test your cameras on your headless device. You could of course also attach a monitor, but i find this more convenient. {: .notice}

sudo raspi-config

-- interface otions, enable vnc

connect through vnc viewer

Install vnc viewer on mac.
Use airaspi.local:5900 as address.

working docker-compose for frigate

Start this as a custom template in portainer.

Important: you need to change the paths to your own paths {: .notice}

version: "3.9"
services:
  frigate:
    container_name: frigate
    privileged: true # this may not be necessary for all setups
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable
    shm_size: "64mb" # update for your cameras based on calculation above
    devices:
      - /dev/apex_0:/dev/apex_0 # passes a PCIe Coral, follow driver instructions here https://coral.ai/docs/m2/get-started/#2a-on-linux

    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /home/aron/frigate/config.yml:/config/config.yml # replace with your config file
      - /home/aron/frigate/storage:/media/frigate # replace with your storage directory
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "5000:5000"
      - "8554:8554" # RTSP feeds
      - "8555:8555/tcp" # WebRTC over tcp
      - "8555:8555/udp" # WebRTC over udp
    environment:
      FRIGATE_RTSP_PASSWORD: "******"

Working frigate config file

Frigate wants this file wherever you specified earlier that it will be.
This is necessary just once. Afterwards, you will be able to change the config in the gui. {: .notice}

mqtt:
  enabled: False

detectors:
  cpu1:
    type: cpu
    num_threads: 3
  coral_pci:
    type: edgetpu
    device: pci

cameras:
  cam1: # <++++++ Name the camera
    ffmpeg:
      hwaccel_args: preset-rpi-64-h264
      inputs:
        - path: rtsp://192.168.1.58:8900/cam1
          roles:
            - detect
  cam2: # <++++++ Name the camera
    ffmpeg:
      hwaccel_args: preset-rpi-64-h264
      inputs:
        - path: rtsp://192.168.1.58:8900/cam2
          roles:
            - detect
    detect:
      enabled: True # <+++- disable detection until you have a working camera feed
      width: 1280 # <+++- update for your camera's resolution
      height: 720 # <+++- update for your camera's resolution

mediamtx

install mediamtx, do not use the docker version, it will be painful

double check the chip architecture here, caused me some headache {: .notice}

mkdir mediamtx
cd mediamtx
wget https://github.com/bluenviron/mediamtx/releases/download/v1.5.0/mediamtx_v1.5.0_linux_arm64v8.tar.gz

tar xzvf mediamtx_v1.5.0_linux_arm64v8.tar.gz && rm mediamtx_v1.5.0_linux_arm64v8.tar.gz

edit the mediamtx.yml file

working paths section in mediamtx.yml

paths:
 cam1:
   runOnInit: bash -c 'rpicam-vid -t 0 --camera 0 --nopreview --codec yuv420 --width 1280 --height 720 --inline --listen -o - | ffmpeg -f rawvideo -pix_fmt yuv420p -s:v 1280x720 -i /dev/stdin -c:v libx264 -preset ultrafast -tune zerolatency -f rtsp rtsp://localhost:$RTSP_PORT/$MTX_PATH'
   runOnInitRestart: yes
 cam2:
   runOnInit: bash -c 'rpicam-vid -t 0 --camera 1 --nopreview --codec yuv420 --width 1280 --height 720 --inline --listen -o - | ffmpeg -f rawvideo -pix_fmt yuv420p -s:v 1280x720 -i /dev/stdin -c:v libx264 -preset ultrafast -tune zerolatency -f rtsp rtsp://localhost:$RTSP_PORT/$MTX_PATH'
   runOnInitRestart: yes

also change rtspAddress: :8554
to rtspAddress: :8900
Otherwise there is a conflict with frigate.

With this, you should be able to start mediamtx.

./mediamtx

If there is no error, you can verify your stream through vlc under rtsp://airaspi.local:8900/cam1 (default would be 8554, but we changed it in the config file)

Current Status

I get working streams from both cameras, sending them out at 30fps at 720p. frigate, however limits the display fps to 5, which is depressing to watch, especially since the tpu doesnt even break a little sweat.

Frigate claime that the TPU is good for up to 10 cameras, so there is headroom.

The stram is completely errant and drops frames left and right. I have sometimes seen detect fps of 0.2, but the TPU speed should definitely not be the bottleneck here. Maybe attach the cameras to a separate device and stream from there?

The biggest issue here is that the google folx seems to have abandoned the coral, even though they just released a new piece of hardware for it. Their most RECENT python build is 3.9. Specifically, pycoral seems to be the problem there. without a decent update, I will be confined to debian 10, with python 3.7.3. That sucks. There are custom wheels, but nothing that seems plug and play.

About the rest of this setup: The decision to go for m.2 E key to save money, instead of spending more on the usb version was a huge mistake. Please do yourself a favor and spend the extra 40 bucks. Technically, its probably faster and better with continuous operation, but i have yet to feel the benefit of that.

TODOs