Object Detection with YOLOv8 and MQTT

This script uses a YOLOv8 model to perform object detection on RTSP camera streams. It then publishes the detection results to an MQTT broker, with support for Home Assistant MQTT discovery.

Features

• Object detection using YOLOv8 from RTSP camera streams.
• Publishes detection results to an MQTT broker.
• Home Assistant MQTT discovery for automatic integration.
• Creates a single sensor per camera that dynamically displays detected objects.
• Saves frames with detections to a specified folder.

Home Assistant Integration

This script uses MQTT discovery to automatically integrate with Home Assistant. For each camera, it creates a single sensor that publishes its state and attributes in a single JSON payload.

• Sensor: A sensor is created for each camera (e.g., sensor.camera1_detections).
  • The state of the sensor is a comma-separated string of the names of the detected objects (e.g., "person, car"). If no objects are detected, the state will be "none".
  • The sensor's attributes contain the full JSON payload, which includes a detections array (with object and confidence for each detection) and a count of the detected objects.

Setup

1. Install dependencies:

   pip install -r requirements.txt

2. **Configuration:

   Create a config.json file in the config directory.

   {
       "cameras": [
           {
               "name": "front_door",
               "url": "rtsp://user:[email protected]/stream1",
               "output_folder": "captures/front_door"
           }
       ],
       "model_path": "yolov8n.pt",
       "interval": 10,
       "mqtt": {
           "broker": "192.168.1.100",
           "port": 1883,
           "user": "mqtt_user",
           "pass": "mqtt_password"
       }
   }

3. Download a YOLOv8 model:

   Download a pre-trained YOLOv8 model (e.g., yolov8n.pt) and place it in the project directory, or specify a full path in config.json.

Usage

Run the script:

python cv_camera.py