This repo contains beta features there are destined for the official integration for Sighthound.

Home Assistant custom integration for people & vehicle detection (with numberplate) using Sighthound Cloud. To use Sighthound Cloud you must register with Sighthound to get an api key. The Sighthound Developer tier (free for non-commercial use) allows 5000 requests per month. If you need more requests per month you will need to sign up for a production account (i.e. Basic or Pro account).

This component adds a pair of image processing entities - one for person detection and one for vehicle detection. The state of the entity is the number of people/vehicles detected in an image.

If save_file_folder is configured, on each new detection an annotated image with the name sighthound_latest.jpg is saved in the configured folder if it doesn't already exist, and over-written if it does exist. The sighthound_latest.jpg image shows the bounding box around detected people/vehicles and can be displayed on the Home Assistant front end using a local_file camera, and used in notifications. Additionally, if save_timestamped_file is configured as True then an image file is created of the processed image, where the file name includes the time of detection.

For each person detected, an sighthound.person_detected event is fired. The event data includes the entity_id of the image processing entity firing the event, and the bounding box around the detected person. For each vehicle detected, an sighthound.vehicle_detected event is fired, with example data below:

{
"event_type": "sighthound.vehicle_detected",
"data": {
    "entity_id": "image_processing.sighthound_vehicle_local_file_1",
    "plate": "CV67CBU",
    "vehicle_type": "car",
    "make": "Ford",
    "model": "Ranger",
    "color": "black",
    "region": "UK"
}

Note that in order to prevent accidentally using up your requests to Sighthound, by default the component will not automatically scan images, but requires you to call the image_processing.scan service e.g. using an automation triggered by motion.

Place the custom_components folder in your configuration directory (or add its contents to an existing custom_components folder). Add to your Home-Assistant config:

image_processing:
  - platform: sighthound
    api_key: your_api_key
    save_file_folder: /config/www/
    save_timestamped_file: True
    always_save_latest_jpg: True
    source:
      - entity_id: camera.local_file

Configuration variables:

• api_key: Your developer api key.
• account_type: (Optional, default dev for Developer) If you have a paid account, used prod.
• save_file_folder: (Optional) The folder to save processed images to. Note that folder path should be added to whitelist_external_dirs
• save_timestamped_file: (Optional, default False, requires save_file_folder to be configured) Save the processed image with the time of detection in the filename.
• always_save_latest_jpg: (Optional, default False, requires save_file_folder to be configured) Always save the last processed image, no matter there were detections or not.
• source: Must be a camera.

It is easy to display the sighthound_latest.jpg image with a local_file camera. An example configuration is:

camera:
  - platform: local_file
    file_path: /config/www/sighthound_latest.jpg
    name: sighthound

Using a counter an automation can be used to count the number of people seen. In configuration.yaml:

counter:
  people_counter:
    name: People
    icon: mdi:alert

In automations.yaml:

- id: 'peoplecounterautomation'
  alias: People Counting Automation
  trigger:
    platform: event
    event_type: sighthound.person_detected
    event_data:
      entity_id: image_processing.sighthound_local_file
  action:
    service: counter.increment
    entity_id: counter.people_counter

The counter is incremented each time a person is detected. The bounding box can in principle be used to include/exclude people based on their location in the image. TODO: add example of using bounding box.

The bounding boxes are formatted to be consumed by the image_processing.draw_box() function. The formatting convention is identical to that used by Tensorflow, where the bounding box is defined by the tuple (y_min, x_min, y_max, x_max) where the coordinates are floats in the range [0.0, 1.0] and relative to the width and height of the image.