Cobani

A bunch of simple scripts to run ML models against your own webcams.

Requirements

Python 2.x, Cobani uses a bunch of libraries, to install them simply run:

pip install -r requirements.txt

Note you don't need all the libraries installed, for example only --training and --predict require tensorflow and tensorflow_hub.

Usage

• Edit the .cobani file to reflect your setup. In my case I'm using the Nest cameras, so you'll need to enter your Nest “access token”.

Fetching images from cameras

• python app.py --picamera --repeat 30 Will fetch images from the Raspberry Pi camera every 30 seconds, and stores them in a folder under images/all/.

• python app.py --nest --repeat 30 Will fetch images (every 30 seconds) from all the Nest cameras associated with your Nest token that's entered in your .cobani settings file. Images are stored in a folder under images/all/.

Train machine learning model with your camera images

• python app.py --train Will train a model based on the images found under images/labeled/. Simply place your own labeled images as subdirectories.

  For example:

  images/labeled/
    ├── door-packages
    │   ├── 1526751023.1088014.jpg
    │   ...
    ├── car
    │   ├── 1526751127.270875.jpg
    │   ...
    ├── mailman
    │   ├── 1526751127.270875.jpg
    │   ...
  

  Make sure to add at least a hundred or more images to be able to learn from them. Also try to have a similar number of images for all categories.

• python app.py --predict --repeat 90 Runs the previously trained machine learning model with your own data. It will read all the camera sources available, run the model and store the results in the predictions/ folder. This is heavy to run on a Raspberry Pi Zero so use the --repeat flag with at least 90 seconds between runs or more.

Slack bots to interact with the camera detections

• python app.py --slack Starts a Slack bot that answers questions about your house. See the plugins section to extend it's capabilities.

  1. To configure your slack bot go to https://api.slack.com/bot-users
  2. Add the token to your .cobani file

• python app.py --slack_changes --repeat 30 Tracks changes on the webcams by looking at the files stored on the predictions/ folder. This is very cheap to run since it only looks at the files on disk.

There's a convenience script called start.sh that runs all the commands at once. This is useful if you want to run that script when your Raspberry Pi boots.

---