This is the source code for a Lego+Raspberry Pi-powered analog film roll scanner. Watch it in action:

You'll need these items to build RoboScan:

• A digital camera with a macro lens: must be compatible with libgphoto2 with image capture and preview support.
• A Raspberry Pi: you may choose a Pi 4 if your camera supports USB 3, otherwise a Pi 2 or 3 is fine.
• A 28BYJ-48 Stepper Motor with ULN2003A driver: easy to find and cheap (about $6)
• 3D-print an adapter to integrate the stepper motor in the legos: use the stepper mount and axis adapter provided by this project (you'll need some bolts to attach the motor to the adapter): https://create.arduino.cc/projecthub/fredrikstridsman/lego-stepperbot-df26b9.
• Adafruit White LED Backlight Module.
• A LED driver such as Recom Power RCD-24-0.70/PL/B or Sparkfun FemtoBuck LED Driver.
• A 50V, 47 μF capacitor
• A high-power LED, such as New Energy LST1-01G03-4095-01: a 4000K white LED, with a CRI (Color Rendering Index) of 95.
• Build the lego part: https://www.mecabricks.com/en/models/r121kn4gvlB.

(made using Fritzing with the help of parts from e-radionica.com and Blomquist)

Put a 50V, 47 μF capacitor between the LED+ and LED - pins of the driver.

Untested - make sure to solder the jumper that can be closed with a glob of solder to double the output current from 330mA to 660mA.

Follow Raspberry foundation documentation, such as: https://projects.raspberrypi.org/en/projects/raspberry-pi-setting-up. Tip: you can set the hostname of your Raspberry Pi to "piscanner" as it's what's used in this tutorial. You can use raspi-config for this.

The easiest is to use the "convenience script" as described here: https://docs.docker.com/engine/install/debian/#install-using-the-convenience-script.

In a nutshell:

curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh
# Add your user to the docker group
sudo groupadd docker
sudo usermod -aG docker $USER

Full docker-compose installation instructions are available here: https://docs.docker.com/compose/install/#install-compose-on-linux-systems.

On Raspbian, the easiest is to install it through Python's pip:

# Install Python3 and pip
sudo apt install python3 python3-pip
# Install docker-compose
sudo pip install docker-compose

git clone https://github.com/bezineb5/RoboScan.git
cd RoboScan

Now, you will ask docker to build and start the application. This might take a while (30-120 minutes).

cd docker
docker-compose up -d --build

It will start all components and restart them at reboot.

The camera must be connected to the Raspberry Pi via USB. It must be compatible with libgphoto2.

Simply navigate to http://piscanner/ (adjust the hostname to your Raspberry Pi)

You can improve the machine learning inference performance by using a Google Coral Edge TPU USB Accelerator plugged on a USB port of the Raspberry Pi. To do so, you have to change the file src/Dockerfile. Replace:

CMD ["python", "webapp.py", "--destination", "/storage/share", "--archive", "/storage/archive", "--temp", "/storage/tmp"]

by:

CMD ["python", "webapp.py", "-tpu", "--destination", "/storage/share", "--archive", "/storage/archive", "--temp", "/storage/tmp"]

The easiest is to code on you PC and deploy docker containers remotely. To do so, enable remote access to the docker daemon.

# Set the DOCKERHOST variable (only once)
# Adjust the hostname to your Raspberry Pi
export DOCKER_HOST=tcp://piscanner.local:2376 DOCKER_TLS_VERIFY=

# Then deploy as usual
cd docker
docker-compose up -d
cd ..