Tweaking Frigate to get the best results

Frigate uses AI to detect people and other objects in your IP camera streams without sending any of your data or video footage to the cloud.

In this video I show you how I have set up Frigate to reliably detect people, dogs and cats in my camera streams. I've achieved this by tweaking the Frigate configuration file to get the best results.

Here is my full Frigate config.yaml file.

 mqtt:
  host: 192.168.1.3
  user: user
  password: <PASSWORD>
ffmpeg:
  hwaccel_args: 
    - -hwaccel
    - vaapi 
    - -hwaccel_device 
    - /dev/dri/renderD128 
    - -hwaccel_output_format 
    - yuv420p
  input_args: -avoid_negative_ts make_zero -fflags +genpts+discardcorrupt -rtsp_transport tcp -stimeout 5000000 -an

rtmp:
  # Optional: Enable the RTMP stream (default: True)
  enabled: True

detectors:
  coral:
    type: edgetpu
    device: usb

#Global Object Settings
objects:
  track:
    - person
  filters:
    person:
      min_area: 5000
      max_area: 100000

cameras:
  FrontCam:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: <PATHTOSTREAM>
          roles:
            - record
        # Low Resolution Stream
        - path: <PATHTOSTREAM>
          roles:
            - detect
            - rtmp
    detect:
      width: 1280 
      height: 720 
      fps: 24
    snapshots:
      enabled: True
      required_zones:
        - Front_Patio
        - Front_Steps
    record:
      enabled: True
      retain:
        days: 5
      events:
        retain:
          default: 10    
        required_zones:
          - Front_Patio
          - Front_Steps
    mqtt:
      required_zones:
        - Front_Patio
        - Front_Steps
    zones:
      Front_Patio:
        coordinates: 19,22,33,44,55,66,77
        objects:
          - person
          - umbrella
        filters:
          person:
            min_area: 5000
            max_area: 100000
            threshold: 0.7
      Front_Steps:
        coordinates: 19,22,33,44,55,66,77
        objects:
          - person
          - umbrella
        filters:
          person:
            min_area: 5000
            max_area: 100000
            threshold: 0.7
    motion:
      mask:
        - 19,22,33,44,55,66,77
  
  BackCam:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: <PATHTOSTREAM>
          roles:
            - record
        # Low Resolution Stream
        - path: <PATHTOSTREAM>
          roles:
            - detect
            - rtmp
    detect:
      width: 1280 
      height: 720 
      fps: 24
    objects:
      track:
        - person
        - dog
        - cat
      filters:
        person:
          threshold: 0.8
          min_area: 5000
          max_area: 100000
    snapshots:
      enabled: True
    record:
      enabled: True
      retain:
        days: 5
      events:
        retain:
          default: 10
    motion:
      mask:
        - 19,22,33,44,55,66,77


Previous
Previous

Installing Z-Wave JS with Docker and Home Assistant

Next
Next

Installing Zigbee2MQTT with Home Assistant