automl-self-checkout
Self-checkout demo for AutoML Vision + Edge TPU
Requirements
- Raspberry Pi 3
- Touch screen monitor for RasPi
- Document camera with USB
- Coral USB accelerator
- Model file: this repo doesn't include the model file. You need to collect training images with labels and create your own .tflite file. See Cloud AutoML Vision Edge doc to learn the model creation process.
- This demo doesn't require any internet connection while working.
- Sample objects. Kids clothes easily fit the camera view.
Install required software
Set screen resolution
sudo vi /boot/config.txt
Uncomment the following lines and edit the width and height:
framebuffer_width=1024
framebuffer_height=600
Install OpenCV
Install OpenCV for USB camera capture.
pip3 install opencv-python
Uninstall numpy to avoid version confliction with Edge TPU SDK.
pip3 uninstall numpy
Install ImageTk
Install ImageTk as UI lib.
sudo apt-get install -y libcblas-dev libhdf5-dev libhdf5-serial-dev libatlas-base-dev libjasper-dev libqtgui4 libqt4-test
sudo apt-get install python3-pil.imagetk
Disable screen saver
sudo apt-get install xscreensaver
sudo reboot
On the desktop, change the screen saver setting as follows:
Menu > Preferences > Screensaver > [Display Modes] > Mode: Disable Screen Saver
Install Edge TPU SDK
See: https://coral.withgoogle.com/docs/accelerator/get-started/#set-up-on-linux-or-raspberry-pi
Place your TF Lite model on the following path
/home/pi/model.tflite
Test the model with Edge TPU
cd automl-self-checkout
python3 test.py
This should return something like:
[(18, 0.84765625), (0, 0.1015625)]
that means the model recognizes the sample polo shirt image at score of 84.7%.
Configure the labels, prices and images
Edit labels and prices definition in main.py, and the image files and their names in /img folder, according to your model definition.
(Optional) Auto start setting
If you want to set up an auto start of the demo, use the following:
Add a service file:
sudo vi /etc/systemd/system/automl-self-checkout.service
[Unit]
Description=automl-self-checkout demo
[Service]
User=pi
ExecStart=/usr/bin/python3 /home/pi/automl-self-checkout/main.py
WorkingDirectory=/home/pi/automl-self-checkout
Environment=DISPLAY=:0.0
[Install]
WantedBy=graphical.target
Reload systemd:
sudo systemctl daemon-reload
Enable and check the status:
sudo systemctl enable automl-self-checkout.service
sudo systemctl status automl-self-checkout.service
Confirm the service would start:
sudo systemctl start automl-self-checkout.service
Run
To run the app:
cd automl-self-checkout
python3 main.py
To stop the app, type <Alt>+F4 on the UI, or:
killall python3
Known issues
- UI response slows down when you runs the app over some hours: the
ui_updates()function slows down due to unknow reason. Restarting the app solve this.