Help with face recognition

[luke518]

Hi,

I have recently migrated from frigate to Viseron, everything is working nicely except face recognition.

I chose code project AI to do the face recognition as looked like it was the only one that was recently updated. I have trained code project AI explorer with multiple faces taken from snapshots in Viseron and using the code project AI inbuilt face recognition check it seems to work.

But in Viseron when I click on the face icon in the events page I can see that it has no idea about whose face is who and it generally gives me about four different guesses of all different people. I think this must be a configuration problem I would be grateful if someone could point out where I've gone wrong.

All the face recognition snapshots get stored in snapshots/face_recognition. I also have a directory called config/face_recognition which is empty.

Here is my config file

---
# /home/luke/viseron/config/config.yaml

## Define your camera
ffmpeg:
  camera:
    camera_2:
      name: doorbell
      host: !secret camera_1_host
      port: 554
      path: /h264Preview_01_main
      fps: 10
      username: !secret camera_1_username
      password: !secret camera_1_password
      recorder: # Store only events for 14 days
        continuous_recording: false
        events:
          max_age:
            days: 14
      # fps: 15
      # width: 640
      # height: 480
      substream:
        path: /h264Preview_01_sub
        port: 554
        fps: 5

## Motion detection
mog2:
  motion_detector:
    cameras:
      camera_2:  # Uses the substream
        fps: 5
        area: 0.05  # Sensitivity for motion detection (adjust as needed)
        trigger_event_recording: false

## Configure the Coral AI (EdgeTPU) object detector
edgetpu:
  object_detector:
    device: usb  # Use the USB-connected Coral Edge TPU
    cameras:  # Camera-specific configurations for the EdgeTPU object detector
      camera_2:  #  Uses the substream
        fps: 5   # Frames per second to send to the Coral for inference
        scan_on_motion_only: true
        zones:
          - name: drive
            coordinates:
              - x: 0
                y: 352
              - x: 658
                y: 255
              - x: 658
                y: 486
              - x: 0
                y: 486
            labels:
              - label: person
                confidence: 0.60  # Confidence threshold for person detection
                trigger_event_recording: true

## NVR configuration for camera_2 (recording
nvr:
  camera_2:

codeprojectai:
  host: 192.168.4.5
  port: 32168
  face_recognition:
    train: true
    labels:
      - person
    save_unknown_faces: true
    cameras:
      camera_2:
        labels:
          - person

Thanks

Luke

[roflcoopter]

I think the problem is that you have set train: true but the config/face_recognition folder is empty which means that Viseron will delete the trained data and no new training will be done since there are no images.

You should at least create a folder for each person that you want to track in the config/face_recognition, so Viseron knows who to track, and then set train: false if you train the model manually.

Let me know if that helps!