Optimizing Frigate Plate Recognizer for Better License Plate Detection

[federicotravaini]

I'm working on a project to automate a gate opening system using license plate recognition. My setup includes Frigate (Plus) with a camera and the Frigate Plate Recognizer integration (GitHub link), which utilizes Plate Recognizer Cloud API for plate detection.

While the system works well for recognizing vehicles, I'm facing a recurring issue: Frigate often selects images that are excellent for vehicle detection but not optimal for license plate recognition. As a result, Plate Recognizer frequently fails to identify the plate correctly because the image captured does not clearly show the plate.

My Questions:

• Is there a way to modify Frigate’s image selection criteria to prioritize clear license plate shots instead of general vehicle images?
• Has anyone implemented a workaround to filter or improve the selection of images sent to the Frigate Plate Recognizer?
• Would it be possible to analyze multiple frames and only send the best-quality image for license plate recognition?

I am aware that the upcoming Frigate 0.16 version will natively integrate license plate recognition, but currently, there is no available beta that can be installed with Docker, so I am unable to test it.

Any insights, suggestions, or experience with similar setups would be greatly appreciated.

[NickM-27]

1. That's already how it works assuming you have license plate enabled

Worth mentioning that 0.16 dev images are available if this is a priority

[federicotravaini]

I would love to test it even if it's a dev build.
Is it available as a docker image? what's the tag?

[hawkeye217]

Dev builds are available here: /~https://github.com/blakeblackshear/frigate/pkgs/container/frigate

Expect instability and bugs. We are not ready to receive feedback, provide support, or look at formal bug reports on dev builds yet, either.

[federicotravaini]

Sorry for bringing this up again, but I was wondering what "enabling license plate" actually means.
I have car detection enabled and a Frigate+ subscription with a custom Frigate+ model using Google Coral.
I assume this is enough to enable plate detection, or am I missing something?

I couldn't find any documentation that explains this in more detail.

[hawkeye217]

There is incomplete documentation for dev builds available here: https://deploy-preview-16390--frigate-docs.netlify.app/

Additional support for unreleased dev builds is not provided at this time.

[NickM-27]

To enable license plate detection on Frigate+ you also need the license plate object defined in your objects.track list

[federicotravaini]

Additional Details:

I've noticed that sometimes the image processed for license plate recognition is blurry, which leads to recognition failures. Even when the vehicle is stationary, no new images are processed after the initial detection, meaning I don't get a second chance to capture a clearer frame.

To work around this, I attempted to create an automation that disables and then immediately re-enables object detection, hoping that Frigate would recognize the vehicle as a new object and process a fresh image. However, this did not work as expected—Frigate did not reprocess the vehicle.

My questions:

Is there a way to force Frigate to reprocess an object after an initial detection failure?
Could adjusting the stationary objects parameters help achieve this?
Are there other workarounds to ensure a new image is captured and sent for plate recognition if the initial frame is suboptimal?
Would using MQTT or the Frigate API allow me to force a new detection? I was wondering if there is an MQTT command that could help, for example, by removing the object from Frigate's memory so that it is detected again as a new object.
Any guidance on this would be greatly appreciated!

[daoudeddy]

I've been testing frigate 0.16 since it was first merged into the dev branch, and im very happy with the results of the LPR.

EDIT: Sometimes, it's better than Gemini :)

[federicotravaini]

I've been testing frigate 0.16 since it was first merged into the dev branch, and im very happy with the results of the LPR.

EDIT: Sometimes, it's better than Gemini :)

would you be so kind to help me to get it working? It's running and recognizing plates, but it just isn't filling the sub_label field

[hawkeye217]

Support is not currently provided for dev builds. You can read the incomplete dev docs here: https://deploy-preview-16390--frigate-docs.netlify.app/configuration/license_plate_recognition/

[federicotravaini]

After a few days of testing, I finally got License Plate Recognition (LPR) working. The issue was that I hadn't enabled Semantic Search, which resulted in the necessary models not being downloaded.

I'm not sure if this is a documentation issue (perhaps it should specify that Semantic Search must be enabled for LPR to function properly) or if it's a missing feature (maybe the required models should be downloaded regardless of whether Semantic Search is enabled).

Either way, I hope this feedback is useful for improving Frigate 0.16. Thanks for all the work on this project!

[daoudeddy]

After a few days of testing, I finally got License Plate Recognition (LPR) working. The issue was that I hadn't enabled Semantic Search, which resulted in the necessary models not being downloaded.

I'm not sure if this is a documentation issue (perhaps it should specify that Semantic Search must be enabled for LPR to function properly) or if it's a missing feature (maybe the required models should be downloaded regardless of whether Semantic Search is enabled).

Either way, I hope this feedback is useful for improving Frigate 0.16. Thanks for all the work on this project!

I'm sure it was mentioned somewhere but i can't remember where, and since this is an unreleased version of frigate, so most of the thing are not ready yet