frigate LPR MQTT documentation help

[tue-kyndal]

Describe the problem you are having

Hi
I ave setup a dedicated LPR camera using the "Using the Secondary LPR Pipeline (Without Frigate+)" method.

I am trying to code some logic using MQTT in nodeRED.
Mostly to test how it works before getting a proper dedicated lpr camera.

I Using the documentation found in the link below, but im not able to get any usable MQTT topics working at all.

https://docs.frigate.video/configuration/license_plate_recognition/
https://docs.frigate.video/integrations/mqtt

I get a new answer on chatgpt every time i ask.
EG:
Key Frigate LPR Topic Structure:
frigate/your_camera_name/lpr: This topic is typically used for license plate recognition events (captures).
frigate/your_camera_name/lpr/plate: If Frigate is publishing the recognized license plate numbers, it might show up here.
frigate/your_camera_name/lpr/snapshot: This topic might be used for sending license plate snapshots.

Could the documentation be more specific and could someone point me in the right direction, as to the structure of the LPR MQTTT topics

Where are the appropriate topics for know and unknown plates and snapshots for the full frame and for the actual plate hitten?

I have tried all sorts of combinations with events, lpr, your_camera_name etc.

Also The debug info is heavy polluted with a none descriptive run line ever sec or so.
Could the debug be cleaned up to display only lines with actual lpr events.

Example from image debug

Version

0.16.2-4d58206

What browser(s) are you using?

Firefox 144.0.2 (64-bit

Frigate config file

# LPR licenes plate reconition global enable
lpr:
  enabled: true
  device: CPU
  min_area: 1000
  recognition_threshold: 0.70
  enhancement: 4 #0-10
  min_plate_length: 6
  model_size: large
  format: ^[A-Z]{2}[0-9]{5}$     
  match_distance: 2 # Allow 2 character variation in plate matching  
  debug_save_plates: true
  known_plates:
    KIA:
      - BD49569
    TOYOTA:
      - DE27517



  dedicated_lpr_camera:
 #   enabled: true
    type: lpr   # required to use dedicated LPR camera mode
    lpr:
      enabled: true
      enhancement: 3 # optional, enhance the image before trying to recognize characters

    ffmpeg: # add your streams
      inputs:
        - path: 
            rtsp://redacted
          roles:
            - detect

    detect:
      enabled: false # disable Frigate's standard object detection pipeline
      fps: 5 # increase if necessary, though high values may slow down Frigate's enrichments pipeline and use considerable CPU
      width: 1920
      height: 1080

    objects:
      track: [] # required when not using a Frigate+ model for dedicated LPR mode
      filters:
        car: {}
    motion:
      threshold: 30
      contour_area: 60 # use an increased value here to tune out small motion changes
      improve_contrast: false
      mask:
        - 0,0,0.377,0,0.381,0.215,0.458,0.358,0.602,1,0,1
        - 0.745,0.383,1,0.691,1,0,0.751,0
        - 0.376,0,0.751,0,0.747,0.223,0.597,0.178,0.461,0.089,0.378,0.081
    record:
      enabled: false # disable recording if you only want snapshots

    review:
      detections:
        enabled: false

    snapshots:
      enabled: false

Relevant Frigate log output

clip form debug on found plate
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Running plate recognition.
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Saving plates for event dedicated-lpr
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Enhancing contrast for recognition image (level: 3)
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Detected text: BD49569 (average confidence: 0.95, area: 13260 pixels)
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Matched plate BD49569 to BD49569 (similarity: 1.000)
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Matched existing plate event for dedicated LPR camera 1762525637.784448-w67tc8: BD49569
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | Plate comparison - Current: BD49569 (score: 0.872, min_conf: 0.88) vs Previous: BD49569 (score: 0.858, min_conf: 0.98) Metrics - Length: 7 vs 7 (scores: 0.50 vs 0.50), Area: 13260 vs 11128, Avg Conf: 0.95 vs 0.98, Char Conf: 0.95 vs 0.98
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: Writing snapshot for 1762525637.784448-w67tc8, BD49569, 1762525639
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | Found license plate. Bounding box: [1418  581 1572  640]
debug | 2025-11-07 15:27:19 | frigate.data_processing.common.license_plate.mixin | dedicated_lpr_camera: YOLOv9 LPD inference time: 15.88 ms

Relevant go2rtc log output

?

FFprobe output from your camera

?

Frigate stats

No response

Install method

Docker Compose

docker-compose file or Docker CLI command

version: "3.7"

services:
          frigate:
            container_name: frigate-0.16.2
            image: ghcr.io/blakeblackshear/frigate:stable
            #cap_add:
            #  - CAP_PERFMON
            privileged: true
            shm_size: "512mb"
            restart: unless-stopped  # always
            volumes:
              - /etc/localtime:/etc/localtime
              - /root/frigate/config.yml:/config/config.yml:rw    #set rw here
              - /media/SSD/frigate/:/media/frigate
              - /media/SSD/frigate/config:/config
              - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
                target: /tmp/cache
                tmpfs:
                  size: 1000000000
            ports:
              - 5000:5000
              - 1935:1935
            #network_mode: "host"
            #Google TPU EDGE
            devices:
              - /dev/apex_0:/dev/apex_0
            environment:
              FRIGATE_RTSP_PASSWORD: "xxxxx"

Object Detector

Coral

Network connection

Wired

Camera make and model

Dahua

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[NickM-27]

The documentation you linked explains exactly where you get updates for LPR. The topic is https://docs.frigate.video/integrations/mqtt/#frigatetracked_object_update (frigate/tracked_object_update) and the payload will look like

{
  "type": "lpr",
  "id": "1607123955.475377-mxklsc",
  "name": "John's Car",
  "plate": "123ABC",
  "score": 0.95,
  "camera": "driveway_cam",
  "timestamp": 1607123958.748393
}

[tue-kyndal]

Frankly i don't really understand what tracked_object_update has to do with license-plate under events.
In https://docs.frigate.video/configuration/license_plate_recognition/

The documentation mentions:
View MQTT messages for frigate/events to verify detected plates.
I'm trying everything under events...license_plate and lpr etc.?

Where are the debug snapshots, like the images I pasted into the discussion above?

I'm sure you are right, but I'm sorry, but the documentation is highly confusing to me.

Just to show why I'm so confused. Perhaps the documentation could be more clear. I have been struggling with this for days.

[NickM-27]

I'm trying everything under events...license_plate and lpr etc.?

No, it is saying literally the frigate/events topic, which publishes payloads about each tracked object. For cars there you can view the recognized plates, but using frigate/tracked_object_update is much easier for directly watching for LPR results

Where are the debug snapshots, like the images I pasted into the discussion above?

They are saved in the debug folder in the filesystem, there is no such MQTT topic for that.

Just to show why I'm so confused. Perhaps the documentation could be more clear. I have been struggling with this for days.

It seems to me your problem is assuming the documentation is not giving all of the details. The documentation is saying exactly where to find information, but you have assumed that there is more to it than there is.

[tue-kyndal]

sorry. I'm not a software engineer.
But I have been using frigate for years on a proxmox, LCX, docker with coral edge.
I simply don't understand the documentation.

Could you give an example of what topic to subscribe to for the know plate:
known_plates:
KIA:
- BD49569

and for an unknown plate.

Because I honestly have no idea anymore at this point - sorry

for known plate is it:
frigate/tracked_object_update/plate #logic cloud be = BD49569
&
frigate/tracked_object_update/name #logic cloud be = KIA

and as per documentation for unknown -- frigate/events MQTT topic as a sub_label (known) or recognized_license_plate

frigate/events/recognized_license_plate #logic cloud be = XX1234 (unknown)

[NickM-27]

I think you are expecting things to work in a way that they don't. Like I said, you should subscribe to frigate/tracked_object_update and that will give the below payload

{
  "type": "lpr",
  "id": "1607123955.475377-mxklsc",
  "name": "John's Car",
  "plate": "123ABC",
  "score": 0.95,
  "camera": "driveway_cam",
  "timestamp": 1607123958.748393
}

you will know first want to filter by only looking at results if the type is lpr, then if it is a known plate if name is supplied, plate will always be supplied

[hawkeye217]

The AI at https://docs.frigate.video is also likely to be a better help than ChatGPT as it has been trained on Frigate documentation and discussions specifically. Just use the "Ask AI" button at the bottom right.

[deanpm70]

I'm seeing an increasing number of people quoting ChatGPT "solutions". I tried asking it some questions about LPR yesterday (I'm doing the same as the OP - figuring out how to write my own notifications), specifically differences in how I would configure lpr for frigate vs. frigate+. Frankly, the answers I was getting were nonsense. Confidently delivered but nonsense nonetheless.

Time for me to have a look at the tracked object topic :)

[tue-kyndal]

The AI at https://docs.frigate.video is also likely to be a better help than ChatGPT as it has been trained on Frigate documentation and discussions specifically. Just use the "Ask AI" button at the bottom right.

I mentioned ChatGPT, because some answers really can lead you stray ;O) Got me confused for sure.

[tue-kyndal]

Hi all.
Thanks a lot for all the feedback.

This is my feedback in return.
My main issues was understanding:

1: The distinction between known and unknown license plate descriped here. This is just absolute Greek to me ;O).

Known/unknown Its not part of the MQTT message -and I got lost in the primary and secondary LPD methods and mqtt events,and when you have known and unknown plate info?

{
  "type": "lpr",
  "id": "1607123955.475377-mxklsc",
  "name": "John's Car",
  "plate": "123ABC",
  "score": 0.95,
  "camera": "driveway_cam",
  "timestamp": 1607123958.748393
}

I fixed it in node-red making my own lookup table, not using the one in my frigate config!
I have no idea how that would be used at this point. perhaps with frigate+??

// Define known plates using a Map for faster lookup and associated names
const knownPlates = new Map([
    ['BD49569', 'Kia'],
    ['DE27517', 'Toyota']
]);

2: I also have don't understand the issue regarding the missing Native license_plate Detecting Model.
Using primary flow gives this error

LprPrimaryCamera is configured to track ['license_plate'] objects, which are not supported by the current model.

Documentation:
Using a Frigate+ (or Native license_plate Detecting) Model
Users running a Frigate+ model (or any model that natively detects license_plate) can take advantage of license_plate detection. This allows license plates to be treated as standard objects in dedicated LPR mode, meaning that alerts, detections, snapshots, and other Frigate features work as usual, and plates are detected efficiently through your configured object detector.

How do you install a Native license_plate Detecting Model ???

I will not go with frigate +
I have no use-case. I'm just playing around and would like to learn and figure out how it works.

Anyway, I ended up Using the Secondary LPR Pipeline (Without Frigate+), on the high Resolution primary stream on the drive way camera.
Then used a snapshot from a secondary stream of the same camera set up normally for frigate.
There are some timing issue, and a delay is needed to ensure the snapshot is available etc.

For others having trouble understanding the documentation, Here is how use the frigate MQTT data in node-red.

pushover high and low priority for unknow/know car in driveway.
snaps are wrong - and I never got a snap out of the dedicated lpr camera! only the backup standard driveway!

this is from a test with the data push button - but live detection is ok in daylight!

flows.json

flow is attached - hope all needed bits are there - its a clip of a much larger flow.

[tue-kyndal]

If any one can figure out how to get the a correctly timed snapshot (with car and plate) out of a e dedicated lpr camera, using the Secondary LPR Pipeline (Without Frigate+), please let me know.

[hawkeye217]

Because you're using the secondary pipeline, snapshots are not created when plates are detected, so there's no way to guarantee an accurately timed snapshot for any plates.

Related to your comments about the documentation above, "known plates" are referring to anything you've specified in known_plates in your LPR config. See the Matching section in the official docs: https://docs.frigate.video/configuration/license_plate_recognition/#matching

[tue-kyndal]

Because you're using the secondary pipeline, snapshots are not created when plates are detected, so there's no way to guarantee an accurately timed snapshot for any plates.

Related to your comments about the documentation above, "known plates" are referring to anything you've specified in known_plates in your LPR config. See the Matching section in the official docs: https://docs.frigate.video/configuration/license_plate_recognition/#matching

I know. But its not clear how to utilize the known/unknown info - see the documentation - its nonsense/Greek to me!

As you can see, I did my own filter in node-red, because I could not figure out the plate category known/unknown?

This discussion is about understanding the documentation! And feedback from your readers how they interpret it

# LPR licenes plate reconition global enable
lpr:
  enabled: true
  device: CPU
  min_area: 1000
  recognition_threshold: 0.70
  enhancement: 4 #0-10
  min_plate_length: 6
  model_size: large
  format: ^[A-Z]{2}[0-9]{5}$ # Standard dk plates that are two letters, followed followed by 5 numbers - space are to small to count!!
  match_distance: 3          # Allow 3 character variation in plate matching due creative new plates and foreign plates etc.
  debug_save_plates: true
  known_plates:
    KIA:
      - BD49569
    TOYOTA:
      - DE27517

[hawkeye217]

In frigate/tracked_object_update, known plates will have a name and a plate. Unknown plates will have a plate and show "None" (or null or undefined, depending on your MQTT client).

In this example, this is a known plate, John's Car:

{
  "type": "lpr",
  "id": "1607123955.475377-mxklsc",
  "name": "John's Car",
  "plate": "123ABC",
  "score": 0.95,
  "camera": "driveway_cam",
  "timestamp": 1607123958.748393
}

[tue-kyndal]

ahh
so in my small node-red example, I could have filtered on "name", if populated (with Kia or Toyota etc.), the plate is known, and if empty or "None" etc, it would be unknown, and used this embedded info for the logic behind the pushover priority level 0 or 1

I see.

Thanks

[hawkeye217]

That's correct.

I'll update the documentation to make this clearer.