Local network image processing using the Google Coral USB accelerator stick exposed via a Flask app -> requires coral-pi-rest-server. The accelerator stick and Home Assistant can be running on different computers as communcation is via the RESTful edndpoints that the Flask app exposes. This code adds an image processing entity in Home Assistant whch has a state that is the number of target objects identified in a camera image, e.g an image with 2 people in it will have the state 2. It will be necessary to experiment with the confidence threshold (a percentage, %) that objects are identified at.

Place the custom_components folder in your configuration directory (or add its contents to an existing custom_components folder). In your configuration.yaml add:

image_processing:
  - platform: google_coral
    ip_address: 192.168.1.107 # the ip of the machine running the flask app
    port: 5000
    confidence: 25 # default 80
    target_object: car # default person
    source:
      - entity_id: camera.local_file
        name: google_coral_car