Defense against fake or stolen "known" license plates triggering automations? LPR + Object Classification?

[dabibbit]

Set-up: You have automations that run based on known good plates. For example, opening a gate or garage door when LPR sees a certain car's plates.

Exploit: An attacker knows this, so they either mock up the "known" plates and use them, or steal the known plate when they know they have time to accomplish their goals before someone returns to the vehicle.

Question: Is there any way to defend against this, other than not having LPR trigger automations that potentially compromise physical security? For example, would requiring Object Classification of either a specific car or a specific make/model/color of car + the known license plate work? Or, would only training OC on a specific car when it also has a clearly visible license plate be sufficient?

Note: Security is a spectrum, and if they're going to both fake your plate and clone your car, I suspect they'll succeed, but I'm more interested in the case where they clone/steal the plate but use whatever vehicle they have handy.

I also realize that triggering the automation would need to be sorted out over in Home Assistant, I'm just trying to figure out the scope of what Frigate can hand it to help defend against this particular exploit.

[inkeep-ai-assistant[bot]]

---

This is a thoughtful security concern. Based on the available documentation, Frigate's LPR system doesn't have built-in mechanisms specifically designed to pair license plate recognition with vehicle appearance verification.

What Frigate LPR provides:

LPR recognizes plates and adds them as a sub_label (for known plates) or to the recognized_license_plate field (for unknown plates) on tracked car or motorcycle objects. (1) The system publishes this data via MQTT on the frigate/events and frigate/tracked_object_update topics with name (if known) and plate. (1)

Current limitations for your use case:

Frigate does not natively support conditional matching like "only trigger if plate X is on a blue sedan." The known_plates configuration maps plate patterns to names, but there's no mechanism to associate a plate with specific vehicle characteristics. (1)

Potential approaches (with caveats):

• Frigate+ models can detect additional object types and may provide more detailed vehicle classification, but the documentation doesn't describe pairing this with LPR for conditional automation logic. (1)

• You could potentially use generative AI descriptions (if enabled) to get vehicle descriptions and handle the matching logic in Home Assistant, but this would require custom automation logic outside Frigate.

• Face recognition is available as a separate enrichment, but that's for identifying people, not vehicles. (2)

The multi-factor verification you're describing (plate + vehicle appearance) would need to be implemented in your automation platform by combining multiple data points from Frigate's MQTT output rather than being a native Frigate feature.

📚 Sources:

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

---

Have another question? Just tag @inkeep.