I'm trying to understand why LPR doesn't run on stationary cars?

[cpw]

I have a vehicle parked in my driveway. It's usually there, it's mine and I've added it to my "recognized plates" list.

It's never been detected by the LPR system because you don't run LPR for stationary cars.

I'm trying to understand why? Is there a reason it behaves like this?

I was hoping to use the LPR system to do basic vehicle presence detection, but this restriction means that any restarts to frigate result in the LPR data being lost and the car becoming "unknown" again.

[inkeep-ai-assistant[bot]]

---

Frigate's LPR is designed to work on moving vehicles, not stationary ones(1). The documentation states: "For moving vehicles, Frigate continuously refines the recognition process, keeping the most confident result. However, LPR does not run on stationary vehicles."(1)

The reason for this design is that LPR is optimized for real-time detection and recognition during motion events(1). When a car or motorcycle is stationary, Frigate treats it as a stationary object and does not run the recognition process(2). This is confirmed in the GitHub issue logs where you see repeated messages like "Not a processing license plate for a stationary car/motorcycle object"(2).

For your use case (vehicle presence detection with a car that's usually parked), this limitation means:

• The plate will only be recognized when the vehicle is moving (pulling in/out of the driveway)(1)
• After a Frigate restart, a stationary vehicle won't be re-recognized until it moves again
• The LPR system isn't designed for persistent presence detection of parked vehicles

This is an architectural decision in how Frigate processes objects. The system focuses on detecting and recognizing plates during motion events rather than continuously monitoring stationary vehicles(1)(2).

📚 Sources:

1. License Plate Recognition (LPR)
2. Recognized Plate senzor in HAS for non Frigate+ native licence plate model

---

Have another question? Just tag @inkeep.

[cpw]

OK, so basically I can't do what I want to do, and it seems pretty much like it's going to stay that way, and I'm going to keep losing tracking unless I move my car every time I restart frigate (a lot at the minute, I'm trying to get it all working). Seems like a fairly significant waste of fuel... 🤔

[hawkeye217]

Even though the LPR models themselves are relatively lightweight, running them continuously on stationary vehicles would still waste resources. The reason is that once a vehicle stops moving, its image (and the license plate region) doesn’t change between frames. Continuously reprocessing that same region provides no new information but still consumes inference time and bandwidth in the detection pipeline.

Frigate is optimized around motion-based events, so it only runs LPR when there’s something new to learn - typically when a vehicle is entering, exiting, or otherwise changing position. This avoids unnecessary inference calls, keeps the pipeline efficient, and ensures resources stay focused on detecting new or changing objects rather than reanalyzing static scenes.

Once you are all set up and configured, I suspect you won't be restarting Frigate often - so this would be a non-issue.

[Anc0dia]

Just a thought, but once a car object is considered stationary run the LPR on that stationary car object. If # is detected, don't run it again for that stationary object, if nothing detected after x time/detection attempts stop LPR detection on that stationary car object.

[hawkeye217]

I'm actually currently working on a solution like that, it should improve the OP's case and others. It should be a part of Frigate 0.17.

[hawkeye217]

I've put up a PR to improve this for the next version of Frigate: #20444