License Plate recognition - does not ever work for me

[dwkirw]

Describe the problem you are having

License plate recognition has never worked once, despite it being turned on and using a 4K stream for detection.
Any ideas what could be wrong?

Version

0.16.1-e664cb2

Frigate config file

mqtt:
  user: ...
  password: ...
  host: 10.0.0.12
  enabled: true
  client_id: frigate_vm

detectors:
  onnx:
    type: onnx
    device: GPU
    num_threads: 3

ffmpeg:
  hwaccel_args: preset-nvidia

birdseye:
  enabled: true
  mode: continuous
  quality: 8
  inactivity_threshold: 45
  layout:
    scaling_factor: 1.0
  width: 2560
  height: 1440
   
auth:
  reset_admin_password: false

ui:
  timezone: Australia/Brisbane

model:
  path: plus://....

audio:
  enabled: true
  listen:
    - bark
    - fire_alarm
    - smoke_detector
    - scream
    - speech
    - yell
    - gunshot

genai:
  enabled: false
  provider: gemini
  api_key: ...
  model: gemini-1.5-flash

semantic_search:
  enabled: true
  reindex: false
  model_size: large


record:
  enabled: true
  retain:
    days: 21
    mode: motion
  alerts:
    retain:
      days: 90
    post_capture: 15
    pre_capture: 15
  detections:
    retain:
      days: 45
      mode: motion
    post_capture: 15
    pre_capture: 15

  sync_recordings: true

detect:
  width: 640
  height: 640
  fps: 3
  enabled: true

objects:
  filters:
    dog:
      min_score: .7
      threshold: .9
    cat:
      min_score: .65
      threshold: .8
    face:
      min_score: .7
    package:
      min_score: .65
      threshold: .9
    license_plate:
      min_score: .6
      threshold: .65
    person:
      min_score: .82
      threshold: .85

go2rtc:
  log:
    format: text
    exec: trace
  streams:

    camerafrontlawn:
      - rtsp://...:[email protected]:554/h265Preview_01_main#tcp
      - ffmpeg:camerafrontlawn#audio=aac
    camerafrontlawn_sub:
      - rtsp://...:[email protected]:554/h264Preview_01_sub#tcp
      - ffmpeg:camerafrontlawn_sub#audio=aac


cameras:
  camera_front_lawn:
    enabled: true
    detect:
      width: 3840    
      height: 2160   
    ffmpeg:
      inputs:

        - path: rtsp://localhost:8554/camerafrontlawn
          input_args:
            - -rtsp_transport
            - tcp
            - -timeout
            - '5000000' # 5s
            - -probesize
            - 512k
            - -analyzeduration
            - '6000000' # 6s
          roles: [record, detect, audio]

        - path: rtsp://localhost:8554/camerafrontlawn_sub
          input_args:
            - -rtsp_transport
            - tcp
            - -timeout
            - '5000000' # 5s
            - -probesize
            - 256k
            - -analyzeduration
            - '4000000' # 4s
            - -fflags
            - +genpts
            - -use_wallclock_as_timestamps
            - '1'
          roles: []

    objects:
      track: [person, dog, cat, car, face, license_plate]
      filters:
        car:
          mask:
            - 0.053,0.142,0.046,0.21,0.11,0.204,0.118,0.131
            - 0.958,0.379,0.9,0.819,0.575,0.519,0.595,0.294,0.708,0.248
        person:
          mask:
            - 0.52,0.221,0.421,0.259,0.305,0.265,0.163,0.293,0,0.286,0.002,0.172,0.005,0.152,0.15,0.149,0.322,0.139,0.419,0.118,0.519,0.11
            - 1,0.613,0.941,0.735,1,0.905
    live:
      streams:
        Sub: camerafrontlawn_sub
        Main: camerafrontlawn

    review:
      alerts:
        labels:
          - person
          - dog
          - car
          - cat
          - rodent
          - license_plate

    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: true
      mask:
        - 0.35,0.048,0.351,0.001,0,0,0,0.044
        - 0,0.349,0.075,0.336,0.119,0.37,0.18,0.398,0.229,0.527,0.247,0.583,0.259,0.64,0.27,0.657,0.287,0.828,0.348,1,0.517,0.99,0.222,1,0.141,0.998,0,1
        - 0.847,0.001,1,0.106,1,0
    audio:
      enabled: true
      listen: [bark, fire_alarm, scream, yell, gunshot]

    snapshots:
      enabled: true
      timestamp: true
      bounding_box: true
      required_zones: []
      retain:
        default: 10



version: 0.16-0

camera_groups:
  Birdseye:
    order: 1
    icon: LuSwitchCamera
    cameras: birdseye

notifications:
  enabled: true
  email: ,,,
face_recognition:
  enabled: true
  model_size: large
lpr:
  enabled: true
classification:
  bird:
    enabled: false

Relevant Frigate log output

nothing relevant

Relevant go2rtc log output

nothing relevant

Frigate stats

nothing relevant

Operating system

Other Linux

Install method

Docker Compose

docker-compose file or Docker CLI command

not relevant

Object Detector

TensorRT

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

Based on your configuration and the LPR documentation, there are several potential issues causing license plate recognition to not work:

Resolution Mismatch Issue
Your detect stream is configured with a 4K resolution (3840x2160) but your global detect settings specify 640x640(1). For LPR to work effectively, the license plate needs to be large enough in the detection frame. With this resolution mismatch, plates may appear too small for recognition(2).

Missing LPR Configuration Parameters
Your LPR configuration only has enabled: true, but you're missing several important parameters that could improve detection(2):

lpr:
  enabled: true
  device: GPU  # Since you're using GPU for detection
  min_area: 1000  # Default minimum area for plate detection
  recognition_threshold: 0.9  # Default recognition threshold

(2)

Model Requirements
Since you're using a Frigate+ model (path: plus://....) and have license_plate in your objects to track, this should work correctly. However, ensure your Frigate+ model actually supports license plate detection(2).

Debugging Steps

1. Enable debug logging for LPR by adding this to your configuration(2):

logger:
  default: info
  logs:
    frigate.data_processing.common.license_plate: debug

(2)

2. Enable debug image saving to see what text areas are being detected(2):

lpr:
  enabled: true
  debug_save_plates: true

(2)

Prerequisites Check
Ensure that(2):

• Cars are being detected first (LPR requires a car or motorcycle detection)
• License plates are clearly visible and readable in your camera view
• The plate area meets the minimum size requirement (default 1000 pixels)

Try adjusting the min_area parameter if plates are being detected but not recognized, or lower the recognition_threshold if recognition is too strict(2).

📚 Sources:

1. Full Reference Config
2. License Plate Recognition (LPR)

---

Have another question? Just tag @inkeep.

[NickM-27]

I'd probably ignore most of this, which frigate+ model are you running?

[dwkirw]

Yeah, it does seem a bit random.
model is the onnx,rocm,openvino,rknn,zmq type, nvidia GPU which is picking up everything else right, just not license plates.

[NickM-27]

The first step to figuring out what is not working would be to enable debug logs for lpr module. That will tell us which step is not working. https://docs.frigate.video/configuration/license_plate_recognition#why-isnt-my-license-plate-being-detected-and-recognized

It's easiest to initially debug this by using a test clip which you can do using https://docs.frigate.video/development/contributing#2-modify-your-local-config-file-for-testing

[NickM-27]

[dwkirw]

Yes, that does happen, there are bounding boxes on cars and license plates but thats the end of it.

[NickM-27]

Cool, detect is working then, proceed with the previous advice

[dwkirw]

So this does seem to be working now - thank you to all that helped.
I didn't change anything except to add this to the lpr settings:

lpr:
enabled: true
device: GPU # Since you're using GPU for detection
min_area: 1000 # Default minimum area for plate detection
recognition_threshold: 0.9 # Default recognition threshold

Maybe it has been working for some time and I hadn't noticed because I was expecting it to show up under a separate 'License Plate' section on the Explore page, but it does not.

[hawkeye217]

Glad it's working for you now! From the docs:

When a plate is recognized, the details are:

• Added as a sub_label (if known) or the recognized_license_plate field (if unknown) to a tracked object.
• Viewable in the Review Item Details pane in Review (sub labels).
• Viewable in the Tracked Object Details pane in Explore (sub labels and recognized license plates).
• Filterable through the More Filters menu in Explore.
• Published via the frigate/events MQTT topic as a sub_label (known) or recognized_license_plate (unknown) for the car or motorcycle tracked object.
• Published via the frigate/tracked_object_update MQTT topic with name (if known) and plate.