[Config Support]: LPR detecting license plates but not reading plate

[repomanz]

Describe the problem you are having

I've recently subscribed to frigate+ and have my configuration working pretty well overall. Although my current configuration is detecting license plates, it's not reading the characters on the plate.

When looking at the video stream I can read the license plate but not in the snapshot that's stored.

I've tried experimenting with enhancement but doesn't seem to have any effect.

Version

.16

Frigate config file

snippets of config:

detect:
  enabled: true

detectors:
  onnx_0:
    type: onnx
    device: gpu
  onnx_1:
    type: onnx
    device: gpu

model:
  path: plus://45e79bc69e9a5188669fcce0ff4e2c00
  #path: /config/yolo_nas_s.onnx
  model_type: yolonas
  width: 640 # <--- should match whatever was set in notebook
  height: 640 # <--- should match whatever was set in notebook
  input_pixel_format: bgr
  input_tensor: nchw
  #labelmap_path: /labelmap/coco-80.txt

lpr:
  enabled: true
  detection_threshold: 0.5
  recognition_threshold: 0.75
  min_area: 1000 # Ignore plates with an area (length x width) smaller than 1500 pixels
  min_plate_length: 4 # Only recognize plates with 4 or more characters
  device: GPU
  model_size: large
  enhancement: 5

objects:
  track:
    - person
    - face
    - dog
    - cat
    - license_plate
    - recognized_license_plate
    - car
    - motorcycle
    - amazon
    - fedex
    - ups
    - bird

cameras:
  garage:
    ffmpeg:
      hwaccel_args: preset-nvidia-h265
      input_args:
        - -avoid_negative_ts
        - make_zero
        - -fflags
        - +genpts+discardcorrupt
        - -probesize
        - 1000M
      output_args:
        record: -f segment -segment_time 10 -segment_format mp4 
          -reset_timestamps 1 -strftime 1 -c:v copy -c:a aac
      inputs:
        - path: 
            rtsp://admin:fvv5n24MC1g6jBfvkqs6c8fyRB@10.180.50.70:554/h265Preview_01_sub
          roles:
            - audio
            - detect
        - path: 
            rtsp://admin:fvv5n24MC1g6jBfvkqs6c8fyRB@10.180.50.70:554/h265Preview_01_main
          roles:
            - record
    birdseye:
      order: 1
    type: lpr
    lpr:
      enabled: true
    detect:
      width: 640
      height: 640
      fps: 5
    objects:
      track:
        - person
        - face
        - cat
        - dog
        - bird
        - bicycle
        - car
        - motorcycle
        - package
        - amazon
        - fedex
        - license_plate
        - recognized_license_plate
    motion:
      threshold: 30
      contour_area: 20
      lightning_threshold: 1.0
      frame_height: 150
      mask:
        - 0.015,0.919,0.363,0.914,0.361,0.961,0.013,0.962
        - 0.724,0.995,0.999,0.999,0.996,0.002,0.839,0,0.811,0.464
      improve_contrast: true
    snapshots:
      enabled: true
      timestamp: false
      bounding_box: true
      retain:
        default: 14
      required_zones:
        - front_driveway
        - front_yard
    record:
      enabled: true
      retain:
        days: 3
        mode: all
      alerts:
        retain:
          days: 7
        pre_capture: 8
        post_capture: 3
      detections:
        retain:
          days: 3
    zones:
      front_driveway:
        coordinates: 
          0.479,0.284,0.669,0.381,0.808,0.46,0.746,0.886,0.721,1,0,1,0,0.326,0.07,0.254,0.164,0.159,0.303,0.218
        inertia: 1
        loitering_time: 0
      front_yard:
        coordinates: 
          0.321,0,0.165,0.157,0.476,0.281,0.677,0.383,0.81,0.457,0.819,0.308,0.835,0
        inertia: 1
        loitering_time: 0
      street:
        coordinates: 0.249,0.065,0.099,0.223,0,0.14,0.136,0
        loitering_time: 0
        speed_threshold: 5
        distances: 28,34,28,34
        inertia: 1

Relevant Frigate log output

No error

Relevant go2rtc log output

no error

Frigate stats

No response

Operating system

UNRAID

Install method

Docker CLI

docker-compose file or Docker CLI command

unraid via frigate app

Object Detector

Coral

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

GPU / Object detection is working. LPR isn't reading license plate characters

[inkeep-ai-assistant[bot]]

Your configuration shows that license plate detection is occurring, but recognition (reading the characters) is not working. Here are the most relevant considerations and troubleshooting steps based on your config and Frigate's documentation:

1. Detection vs. Recognition:
   If you see a bounding box labeled license_plate but no characters are read, it means the detection step is working but the recognition step is failing to extract characters(1).

2. Plate Area and Resolution:
   The min_area is set to 1000, which is low, but if the plates are small in the image, recognition may still fail. Try increasing your camera's detect stream resolution or lowering the camera angle so the plate appears larger in the image(1).

3. Enhancement Setting:
   Excessive enhancement can blur characters, making recognition harder. Try setting enhancement: 0 and test. If the plate is still unreadable, gradually increase it (e.g., 3 or 5), but avoid high values(1).

4. Model Size:
   The large model is more accurate at finding text but not at recognizing it. If your plates are single-line and typical, using small may be faster and just as accurate for reading(1).

5. Recognition Threshold:
   Your recognition_threshold is 0.75 (default is 0.9). Lowering this makes Frigate less strict, but if the image quality is poor, even a low threshold may not help(1).

6. Debugging:
   Set debug_save_plates: True in your lpr config. This will save images of the detected plate regions in /media/frigate/clips/lpr/. Check these images: if you cannot read the plate text in these crops, Frigate will not be able to either. If the crops are poor, adjust the camera position or stream quality(1).

7. Known Plates and Format:
   If your plates follow a strict format, try setting the format option with a regex matching your plate style. This can reduce false positives(1).

8. Objects to Track:
   For Frigate+ models that natively detect license_plate, you should have license_plate in your track list. (You also have recognized_license_plate—this is not a valid object to track and can be removed.)(1)

Example config for debugging:

lpr:
  enabled: true
  debug_save_plates: true
  enhancement: 0  # Start with 0 and adjust if needed
  recognition_threshold: 0.75
  min_area: 1000
  model_size: large

(1)

Summary:

• Set debug_save_plates: true and review the crops.
• Adjust enhancement and check camera/stream quality.
• Remove recognized_license_plate from your track list.
• Increase plate size in the image if possible.

If after these steps the issue persists and the crops look readable to you, there may be an edge case requiring further review.

📚 Sources:

1. License Plate Recognition (LPR)

---

Have another question? Just tag @inkeep.

[NickM-27]

Do you have this camera configured as a dedicated LPR camera? That doesn't seem right. Remember that LPR also does not run on stationary objects

[repomanz]

Even if the vehicle is moving into driveway it won't read the plate. attached camera config

cameras:
garage:
ffmpeg:
hwaccel_args: preset-nvidia-h265
input_args:
- -avoid_negative_ts
- make_zero
- -fflags
- +genpts+discardcorrupt
- -probesize
- 1000M
output_args:
record: -f segment -segment_time 10 -segment_format mp4
-reset_timestamps 1 -strftime 1 -c:v copy -c:a aac
inputs:
- path:
rtsp://admin:fvv5n24MC1g6jBfvkqs6c8fyRB@10.180.50.70:554/h265Preview_01_sub
roles:
- audio
- detect
- path:
rtsp://admin:fvv5n24MC1g6jBfvkqs6c8fyRB@10.180.50.70:554/h265Preview_01_main
roles:
- record
birdseye:
order: 1
type: lpr
lpr:
enabled: true
detect:
width: 640
height: 640
fps: 5
objects:
track:
- person
- face
- cat
- dog
- bird
- bicycle
- car
- motorcycle
- package
- amazon
- fedex
- license_plate
- recognized_license_plate
motion:
threshold: 30
contour_area: 20
lightning_threshold: 1.0
frame_height: 150
mask:
- 0.015,0.919,0.363,0.914,0.361,0.961,0.013,0.962
- 0.724,0.995,0.999,0.999,0.996,0.002,0.839,0,0.811,0.464
improve_contrast: true
snapshots:
enabled: true
timestamp: false
bounding_box: true
retain:
default: 14
required_zones:
- front_driveway
- front_yard
record:
enabled: true
retain:
days: 3
mode: all
alerts:
retain:
days: 7
pre_capture: 8
post_capture: 3
detections:
retain:
days: 3
zones:
front_driveway:
coordinates:
0.479,0.284,0.669,0.381,0.808,0.46,0.746,0.886,0.721,1,0,1,0,0.326,0.07,0.254,0.164,0.159,0.303,0.218
inertia: 1
loitering_time: 0
front_yard:
coordinates:
0.321,0,0.165,0.157,0.476,0.281,0.677,0.383,0.81,0.457,0.819,0.308,0.835,0
inertia: 1
loitering_time: 0
street:
coordinates: 0.249,0.065,0.099,0.223,0,0.14,0.136,0
loitering_time: 0
speed_threshold: 5
distances: 28,34,28,34
inertia: 1

[hawkeye217]

You should follow the official documentation for debugging LPR issues: https://docs.frigate.video/configuration/license_plate_recognition/#how-do-i-debug-lpr-issues