Why face detection or lpr don't work on my config ?

[edouardkleinhans]

Hello,

I'm trying to implement Face detection and LPR on my imou camera but it don't work and I can't find why.

Please find my current configuration:

mqtt:
  enabled: true
  host: 192.168.20.100
  user: frigate
  password: XXXXXXXXX
  port: 1883

ffmpeg:
  hwaccel_args: preset-vaapi

detectors:
  coral:
    type: edgetpu
    device: usb

go2rtc:
  webrtc:
    candidates:
      - 192.168.20.50:8555
      - stun:8555
  streams:
    garage:
      - rtsp://admin:XXXXXXXXX@192.168.20.93:554/cam/realmonitor?channel=1&subtype=0
      - ffmpeg:garage#audio=opus
    garage_sub:
      - rtsp://admin:XXXXXXXXX@192.168.20.93:554/cam/realmonitor?channel=1&subtype=1
      - ffmpeg:garage_sub#audio=opus

    cuisine:
      - rtsp://admin:XXXXXXXXX@192.168.20.91:554/cam/realmonitor?channel=1&subtype=0
      - ffmpeg:cuisine#audio=opus
    cuisine_sub:
      - rtsp://admin:XXXXXXXXX@192.168.20.91:554/cam/realmonitor?channel=1&subtype=1
      - ffmpeg:cuisine_sub#audio=opus


objects:
  track:
    - person
    - cat
    - dog
    - bird
    - face
  filters:
    person:
      min_area: 0.075


#################
### RECORDING ###
#################
record:
  enabled: True
  retain:
    days: 7
    mode: motion
  alerts:
    retain:
      days: 30
  detections:
    retain:
      days: 30

snapshots:
  enabled: True
  retain:
    default: 30

semantic_search:
  enabled: true
  reindex: false
  model_size: small

detect:
  enabled: true

face_recognition:
  enabled: true
  model_size: small
  detection_threshold: 0.5
  
audio:
  enabled: true  # Temporarily disabled due to HEVC stream issues
  max_not_heard: 30
  min_volume: 500
  listen:
    - bark
    - fire_alarm
    - scream
    - speech
    - yell

review:
  alerts:
    labels:
      - person
    required_zones: []
  detections:
    labels:
      - dog
      - cat
      - speech
      - car
      - bird
    required_zones: []

cameras:
  garage:
    enabled: true
    onvif:
      host: 192.168.20.93
      port: 80
      user: admin
      password: XXXXXXX
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac
      inputs:
        - path: rtsp://127.0.0.1:8554/garage
          input_args: preset-rtsp-restream
          hwaccel_args: preset-intel-qsv-h264
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/garage_sub
          input_args: preset-rtsp-restream
          hwaccel_args: preset-intel-qsv-h264
          roles:
            - audio
            - detect
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5
    lpr:
      enabled: true    
    live:
      streams:
        garage_sub: garage_sub
    zones:
      Devant_garage:
        coordinates: 
          0.2,1,0.336,0.378,0.333,0.078,0.711,0.112,0.697,0.41,0.816,1
        loitering_time: 0
  cuisine:
    enabled: true
    onvif:
      host: 192.168.20.91
      port: 80
      user: admin
      password: XXXXXXX
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac
      inputs:
        - path: rtsp://127.0.0.1:8554/cuisine
          input_args: preset-rtsp-restream
          hwaccel_args: preset-intel-qsv-h265
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/cuisine_sub
          input_args: preset-rtsp-restream
          hwaccel_args: preset-intel-qsv-h264
          roles:
            - audio
            - detect
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5
    lpr:
      enabled: False      
    live:
      streams:
        cuisine_sub: cuisine_sub
    zones:
      porte_exterieur:
        coordinates: 0.15,1,0.115,0.351,0.373,0.323,0.392,0.998
        inertia: 3
        loitering_time: 0
      porte_cuisine:
        coordinates: 0.604,0.995,0.609,0.398,0.958,0.384,0.958,0.819
        loitering_time: 0
version: 0.16-0


lpr:
  enabled: true
classification:
  bird:
    enabled: true
camera_groups:
  Ext:
    order: 1
    icon: LuCar
    cameras: garage
  Int:
    order: 2
    icon: LuCctv
    cameras:
      - salon
      - entree
      - cuisine

Is it due to resolution of my stream ? or another missing configuration ?

Thanks

[inkeep-ai-assistant[bot]]

Here are some key points to check based on your configuration and Frigate requirements for Face Detection and License Plate Recognition (LPR):

Face Detection / Face Recognition:

• If you are using a Frigate+ model (or any custom model that natively detects faces), you should have "face" in your tracked objects list. If you are not using such a model, "face" should NOT be in the objects.track list(1).
• Face detection requires that a person is first detected. If "person" is not being detected reliably (due to camera angle, resolution, or other reasons), face detection will not be triggered(1).
• The stream used for detection should have enough resolution and clarity for faces to be detected. The recommended minimum area for a face before recognition runs is 500 pixels. You can adjust the min_area parameter if needed(1).
• Face recognition must be enabled globally (face_recognition.enabled: true). You have this set, which is correct(1).

LPR (License Plate Recognition):

• LPR must be enabled globally (lpr.enabled: true) and can be disabled per camera(2).
• For non-dedicated LPR cameras, your camera must be configured to detect objects of type car or motorcycle. If these objects are not being detected, LPR will not run(2).
• If you are using a Frigate+ model that detects license plates, you should add license_plate to your tracked objects list. If not, do NOT add it(2).
• The detection stream must have enough resolution for license plates to be detected and recognized. The default minimum area for a plate is 1000 pixels (32x32), but you may need to increase this for higher resolution cameras or distant plates(2).
• LPR does not run on stationary vehicles, only moving ones(2).

Potential Issues in Your Config:

• You have "face" in objects.track. If you are NOT using a Frigate+ or face-detecting model, remove "face" from this list(1).
• You do not have "car" or "motorcycle" in objects.track, which is required for LPR to trigger on non-dedicated LPR cameras(2).
• Your detect stream resolution is set to 640x480, which is generally sufficient, but if your camera is far from the subject, recognition may fail. Consider increasing resolution if faces or plates are small(1)(2).
• Ensure motion detection is enabled and working, as both face and LPR rely on motion-triggered detection(2)(1).

Actionable Steps:

1. For face detection, only include "face" in the objects.track list if you have a model that supports it natively(1).
2. For LPR, add "car" and/or "motorcycle" to your objects.track list for the relevant cameras(2).
3. Make sure your detection stream is clear enough and at a sufficient resolution for the task.
4. Check that motion detection is working and not masked or disabled for your cameras.

For more details, review:

• Face Recognition
• License Plate Recognition (LPR)

📚 Sources:

1. Face Recognition
2. License Plate Recognition (LPR)

---

Have another question? Just tag @inkeep.

[edouardkleinhans]

I've tried to disable face on object, but nothing change. Still no face detection

[hawkeye217]

You should enable debug logs for both face recognition and LPR so you can see exactly what Frigate is doing.

Follow the steps in the official documentation:

https://docs.frigate.video/configuration/face_recognition#how-do-i-debug-face-recognition-issues

https://docs.frigate.video/configuration/license_plate_recognition#how-do-i-debug-lpr-issues

[edouardkleinhans]

Still not workin. Don't know why.
I see this error on startup, don't know if it's related :
2025-09-04 11:57:26.042872536 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 522, index: 1, mask: {2, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.

[hawkeye217]

Did you enable debug logs as I suggested? If so, what do they show? There's no way to be able to help without seeing a full copy of the logs.

The pthread message you're seeing is related to running Frigate in Proxmox and Unraid: #15351