Skip to content

Latest commit

 

History

History
201 lines (145 loc) · 5.31 KB

README.md

File metadata and controls

201 lines (145 loc) · 5.31 KB

Garage Door State AI

This project is an AI-powered solution for monitoring and managing the state of a garage door using image classification, MQTT, and integration with Home Assistant.

Features

  • Image Classification:

    • Uses TensorFlow/Keras to classify whether the garage door is open or closed based on uploaded images.

  • Home Assistant Integration:

    • Publishes the door state via MQTT with auto-discovery for seamless integration into Home Assistant.

  • Web Interface:

    • User-friendly interface to upload images for prediction.
    • Live retraining logs displayed during model retraining.

  • Model Retraining:

    • Ability to retrain the model directly from the web interface with dataset updates.
    • Dataset management for categorized (open/closed) images.

  • Automated Cleanup:

    • Removes old images from the dataset and upload directories to optimize disk space.

Table of Contents

  1. Installation
  2. Environment Variables
  3. Usage
  4. Home Assistant Setup
  5. Retraining the Model
  6. MQTT Topics
  7. Contributing
  8. License

Installation

  1. Clone the repository:

    git clone https://github.com/GerryCrooked/garage-door-state.git
cd garage-door-state

  2. Build and start the Docker container:

    docker-compose up -d

  3. Access the web interface at:

    http://<your-server-ip>:5000

Environment Variables

Create a .env file in the root directory and define the following variables:

# MQTT Configuration
MQTT_HOST=<your-mqtt-broker-ip>
MQTT_PORT=1883
MQTT_USER=<your-mqtt-username>
MQTT_PASSWORD=<your-mqtt-password>

# Device Configuration
TOPIC=homeassistant/binary_sensor/garage_door_status
CONFIG_TOPIC=homeassistant/binary_sensor/garage_door_status/config
STATE_TOPIC=homeassistant/binary_sensor/garage_door_status/state
SENSOR_NAME=Garage Gate Open/Close Sensor
DEVICE_ID=garage_gate_status_sensor
DEVICE_NAME=Garage Gate
DEVICE_MODEL=ARES1500
DEVICE_MANUFACTURER=BFT

# Timezone
TZ=Europe/Berlin

# Web Server Port
PORT=5000

Usage

Upload Images

  • Use the web interface to upload images for prediction.
  • The system will classify the image as open or closed and publish the result to the MQTT broker.

View Predictions

  • Visit the /last_prediction page to view the most recent prediction and the associated image.

Manage Dataset

  • Move images to open or closed categories via the /action/<action>/<filename> endpoint.
  • Download or clear the dataset as needed via the web interface.

Home Assistant Setup

This project supports MQTT auto-discovery, so minimal setup is required in Home Assistant. Ensure MQTT is configured correctly in Home Assistant.

The device and binary sensor will automatically appear with the following properties:

  • Device Name: Garage Gate
  • Entity: binary_sensor.garage_gate_status
  • State: open or closed

Manual Configuration (Optional)

If auto-discovery fails, add the following to your configuration.yaml:

binary_sensor:
  - platform: mqtt
    name: "Garage Gate Status"
    state_topic: "homeassistant/binary_sensor/garage_door_status/state"
    payload_on: "open"
    payload_off: "closed"
    device_class: "door"

Automating Camera Snapshot Uploads

Use the following automation to take a snapshot of a camera and upload it to the server:

Home Assistant Automation

automation:
  - alias: Upload Garage Camera Snapshot
    description: Takes a screenshot of the garage camera and uploads it to the server
    trigger:
      - platform: time_pattern
        minutes: "/1"  # Run every minute
    action:
      - service: camera.snapshot
        target:
          entity_id: camera.garage_camera  # Replace with your camera entity
        data:
          filename: /config/www/garage_snapshot.jpg
      - service: shell_command.upload_garage_snapshot

Shell Command in Home Assistant

Add this to your configuration.yaml under the shell_command section:

shell_command:
  upload_garage_snapshot: >
    if [ -f /config/www/garage_snapshot.jpg ]; then
      rm /config/www/garage_snapshot.jpg;
    fi && curl -X POST -F "file=@/config/www/garage_snapshot.jpg" http://<your-server-ip>:5000

Retraining the Model

Automatic

  • Use the /retrain endpoint to start retraining the model.
  • Logs are displayed live on the /last_prediction page.

Manual

  • Add new images to the dataset/open and dataset/closed directories.
  • Run the retraining script: 
    python retrain.py /app/model/garage_door_model.keras

MQTT Topics

Topic Description
homeassistant/binary_sensor/garage_door_status/config Auto-discovery configuration
homeassistant/binary_sensor/garage_door_status/state Current state of the door

Contributing

Contributions are welcome! Please follow these steps:

  1. Fork the repository.
  2. Create a new branch for your feature or fix.
  3. Submit a pull request.

License

This project is licensed under the MIT License. See the LICENSE file for details.