- Update the Raspberry pP's code bij executing the following two lines.
sudo apt update
sudo apt full-upgrade
-
Enable the pi cam by running
sudo raspi-config, and go to the promp in Interfacing Options > Camera. And reboot. -
Execute
./raspberry/install_dependencies.sh. -
Change your work directory to the webapp folder
cd raspberry/webserver/webapp. -
Create an environment file
cp .env.example .env.local. -
Fill in the api uri, this should be the IP of the Raspberry:
http://+ IP +/api/note the trailing slash.- Alternatively you can just do
/api/.
- Alternatively you can just do
-
Execute
npm install && npm run build.
-
Plug the Zumo into the host machine via USB.
-
Open the
zumo/build.shfile and change the port to the port your Zumo uses. -
Make sure that
arduino-cliandscreenare installed. -
Flash the software by running the script
./zumo/build.sh.
-
Plug a beacon into the host machine via USB.
-
Open the
beacon/build.shfile and change the port to the port your beacon uses. -
Make sure that
arduino-cliandscreenare installed. -
Flash the software by running the script
./beacon/build.sh. -
Repeat the last step until all four beacons have beel flashed.
To train the Tensorflow Lite model follow the tutorials from EdjeElectronics:
After this you should follow some steps to convert the model to a EdgeTPU model (runnable for the Google Coral)
- Install
edgetpu-compiler:
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
sudo apt-get update
sudo apt-get install edgetpu-compiler-
Run the command
edgetup_compiler [folder of tflite model]. -
Rename
detect_edgetpu.tflitetoedgetpu.tflite