Overview

Control code and utilities for ScrubCam, an edge-AI-enabled field wildlife camera system.

Hardware

The target hardware for the field device is a Raspberry Pi 4 Model B single board computer with a Picamera-style camera, a Google Coral USB Accelerator, and the AdaFruit PiTFT screen (2.8" resistive touch model with 4 GPIO-connected buttons) attached. Typically several storage devices are connected via the USB ports on the Pi (e.g. uSD card readers). This hardware is typically integrated into a larger assembly that includes a weatherproof enclosure and batteries and--in some variants--a charge controller with external solar panels.

Operating System

Currently, the system runs and has been tested on Raspian Stretch and Buster. Problems have been encountered with Bullseye that are believed to be mostly (if not entirely) linked to picamera no longer being supported and replaced by picamera2. At some point the ScrubCam developers will decide whether to switch the system over to picamera2 or provide instructions here for re-enabling the legacy camera stack.

Setup

Clone the ScrubCam repo

git clone https://github.com/icr-ctl/scrubcam.git

Install and set up virtualenv and virtualenvwrapper

sudo pip3 install virtualenv virtualenvwrapper

echo -e "\n# Virtualenv and virtualenvwrapper stuff" >> ~/.bashrc
echo "export WORKON_HOME=$HOME/.virtualenvs" >> ~/.bashrc
echo "export VIRTUALENVWRAPPER_PYTHON=/usr/bin/python3" >> ~/.bashrc
echo "source /usr/local/bin/virtualenvwrapper.sh" >> ~/.bashrc
source ~/.bashrc

Create virtual environment for scrubcam project

Create a virtual environment named "scrubcam_env":

mkvirtualenv scrubcam_env

Note that ScrubCam requires Python 3 so if the default on your system is Python 2 (it is on Stretch and Buster), make sure the virtual environment will use Python 3 by using this command instead:

mkvirtualenv scrubcam_env -p python3

Activate virtual environment (not necessary if you just made it):

workon scrubcam_env

Install ScrubCam dependencies

Update apt package sources list

sudo apt update

Run installation script

Navigate to inside of the local copy of the scrubcam code repository and run the install script:

cd scrubcam
./install_on_pi.sh

Additionally, the install script will enable the camera, if not already enabled.

Note: this install script is intended for installing on a Pi. If you are installing on your regular computer to access some of the ScrubCam tools that are useful there, we should probably make an install script for you. In the meantime the main thing you want to do is:

pip install .

This will install the required dependencies, which do not include Pi specific dependencies.

PyCoral Dependencies

Scrubcam uses the Google Coral USB Accelerator and the associated packages to interface with this hardware cannot be installed with pip and their installation is not currently included in the install script.

Install PyCoral Packages using included script

Run the following script to install apt packages needed for the PyCoral to work. The script will prompt for a reboot; this is only necessary if you don't have physical access to the system to unplug/replug the Coral USB Accelerator.

./install_pycoral.sh

Link PyCoral packages into virtual environment

To be able to access the PyCoral packages from within your virtual environment you must create symlinks within the virtual environment files to where they are located.

Locate the packages:

dpkg -L python3-pycoral
dpkg -L python3-tflite-runtime

For each of those two packages, find a line similar to "/usr/lib/python3/dist-packages/pycoral"

Go into the folder of the virtual environment you created (we used scrubcam_env above) and create a symlink for each of those packages depending on where they are actually located on your system. For example:

cd ~/.virtualenvs/scrubcam_env/lib/python3.7/site-packages/
ln -s /usr/lib/python3/dist-packages/pycoral
ln -s /usr/lib/python3/dist-packages/tflite_runtime

Note that .virtualenvs is where we set up for virtual environments to go (e.g. those created using mkvirtualenv) but this may not be the case if using venv or virtualenv to create the virtual environment. Replace .virtualenvs with the folder that your virtual environment is saved to. This may be a regular folder in some instances.

Get models for testing everything is working

Install package that includes some models that we can use to test things:

sudo apt install pycoral-examples

Now there should be a bunch of models in /usr/share/pycoral/examples/models. One of these is actually used by the example config file in cfgs folder.

Note: If you did the full set of steps from the PyCoral website above you would have installed one MobileNet image classifier model. There are similar workflows/scripts within that repo to get some of the same models that are in pycoral-examples.

Usage

On Raspberry Pi

The core program in the project is scrubcam.py which runs on the Raspberry Pi which is controlling the field device. It accepts a YAML configuration file as a command line argument. There is an example config file ./cfgs/config.yaml.example

You can copy this example file and modify it to your specific purposes and thus then run ScrubCam on a properly set up system (see Setup section above for how to set up system) via:

Usage:
    ./scrubcam.py [OPTIONS] cfgs/YOUR_CONFIG_FILE.yaml
    
Options:
    -h, --help          Output a usage message and exit
    -c, --continue      Continue saving images to the most recent session on ScrubDash

On remote server: ScrubDash

For monitoring the activity of a ScrubCam there is also a Plotly-Dash-based server/dashboard program called scrubdash.py intended to be run on a machine on the same network (e.g. back in the lab or on a laptop in the field). Scrubdash is used to organize, visualize, and analyze images that are sent by ScrubCams.

conservationtechlab/scrubcam