Cameras

Frigate false alerts: detection tuning

Cut Frigate false positives the right way — min_score vs threshold, object area, zones, and the limits of motion masks.

Problem summary

Frigate false alerts are tuned in layers: motion → object detection → zones → object filters → review (alerts vs detections). The knobs that matter most are min_score vs threshold (they're different), object min/max area, and zones — and a motion mask only suppresses motion-triggering, not an object once it's tracked.

Operator snapshotEvidence first

First proof

Decide whether you're getting false OBJECTS or just noisy motion.

Screen to open

Frigate UI > Settings > Debug (enable zones)

Expected signal

You can tell from the UI whether bogus objects are detected or motion is just busy.

Stop boundary

Don't expect a motion mask to suppress an already-tracked object.

Layer path

1False alerts in Frigate are tuned in layers: motion detection → object detection → zones → object filters → review (alerts vs detections). Each layer has a distinct knob.

2The two filters people confuse: min_score drops any single detection below it; threshold is the score the object's median computed score must reach to count as a real object. They are different and both matter.

3Masks are not magic: a motion mask only suppresses motion-triggering in that region — once an object is tracked it's no longer hidden by the mask.

4Zones define where an object must be to matter; debugging with zones/debug view shows exactly what Frigate sees.

Runbook

Step-by-step runbook

Start here. Do each check in order, compare it to the expected result, and stop when the evidence explains the failure or the safe stop point applies.

Watch the debug view first

Check: Enable Debug with zones and observe the false detections live.

Expected result: You can see scores, box sizes, and zones for the bad hits.

If not: This tells you which layer to tune instead of guessing.

Tune score filters

Check: Raise min_score to drop weak hits; set threshold for the object's median confidence.

Expected result: Weak false objects stop appearing while real ones remain.

If not: Remember min_score and threshold are different gates.

Add area limits

Check: Set min_area/max_area to reject wrong-sized boxes for that camera.

Expected result: Leaf-sized and whole-frame false objects are filtered.

If not: Use pixels or a 0–1 fraction of the frame.

Define zones and require them for alerts

Check: Draw zones for the areas that matter and require in-zone presence.

Expected result: Alerts only fire for objects where you care.

If not: Detections outside zones no longer notify.

Mask perpetual motion, then split alerts vs detections

Check: Add motion masks only for trees/flags/timestamps, and rely on review (alerts vs detections) for notifications.

Expected result: Notification noise drops to real, in-zone, well-scored objects.

If not: Re-evaluate over a full day/night before tightening further.

Decision tree

If: Weak/low-confidence false objects.

Then: min_score too low.

Action: Raise min_score for that object type and re-evaluate over a day.

If: Wrong-sized false objects (tiny/huge).

Then: Missing area filters.

Action: Set min_area/max_area (pixels or a 0–1 fraction of the frame).

If: Real objects, but alerting everywhere.

Then: No/incorrect zones.

Action: Define zones and require in-zone presence for alerts.

If: Constant motion in part of the frame.

Then: Perpetual-motion region (trees/flag/timestamp).

Action: Add a motion mask for that region only — not where you want objects detected.

Safe stop: Don't expect a motion mask to suppress an already-tracked object.

Evidence

Evidence table

Symptom	Evidence to collect	Likely layer	Next action
Lots of low-confidence false hits.	Object scores vs min_score/threshold.	Score filters	Raise min_score; tune threshold.
Tiny or frame-sized false boxes.	Bounding-box area vs min/max area.	Area filters	Set min_area/max_area for the scene.
Alerts from areas you don't care about.	Whether zones are defined and required.	Zones	Draw zones; require in-zone for alerts.
Motion spam from trees/flags/timestamp.	Where motion triggers in the debug view.	Motion mask region	Mask only the perpetual-motion region.

Reference

Commands and settings paths

Use the debug view with zones

Frigate UI > Settings > Debug (enable zones)

Where: In the Frigate web UI on the live camera.

Expected: You see bounding boxes, scores, and zone entry in real time.

Failure means: It reveals whether the issue is score, area, or zone placement.

Safe next step: Tune the matching filter and watch the debug overlay update.

Inspect object scores

Frigate UI > Explore/Events — read the score on false events

Where: Against recent false detections.

Expected: You can see the actual scores the false objects received.

Failure means: Scores near your threshold show where to set min_score/threshold.

Safe next step: Set min_score just above the false-hit band; keep true objects above threshold.

Review the config filters

Frigate config: objects.filters.<label>.min_score / threshold / min_area / max_area; zones

Where: In the Frigate config (validate against current docs).

Expected: Filters and zones reflect your intent for each camera.

Failure means: Schema changes between versions — confirm keys against the live docs.

Safe next step: Apply, restart Frigate, and re-evaluate over a full day/night cycle.

Hardware boundary

Hardware and platform boundary

Change only when

Invest in better camera placement/lighting and a capable detector before over-tuning — clean input beats aggressive filters.

Evidence that matters

A camera with a good substream for detection, sensible field of view, and a detector with headroom (see the Frigate hardware planner).

Evidence that does not matter

Chasing a single 'magic' setting — false-positive control is the combination of score, area, zones, and masks.

Avoid

Over-masking (hides real objects) or treating every detection as an alert (notification fatigue).

Related tool/calculator

Use the linked calculator or tool to turn this runbook into numbers for your exact setup.

Frigate hardware planner

Last reviewed

2026-06-02 · Reviewed by HomeTechOps. Built from June-2026 research verified against the Frigate object-filters, masks, and zones docs; centers the min_score-vs-threshold distinction (which most guides get wrong) and the motion-mask limitation. The alerts-vs-detections (Review) terminology is recent-version-specific — verify against the live docs before publishing exact keys.

Sources/assumptions

Assumes a working Frigate install already detecting objects (if you get no frames, fix that first).
Frigate's filter/zone/review terminology is version-specific — verify keys against the current docs before applying.
Tuning reduces false positives; it can't fully compensate for poor camera placement or lighting.

Source-backed checks

HomeTechOps turns official docs and conservative safety rules into a shorter runbook. These links are the source trail for the page direction.

Frigate docs: Recommended hardwareUsed for detector inference times, the cameras ≈ (1000 ÷ inference_ms) ÷ detect_fps capacity model, and the current 'Coral no longer recommended for new installs' guidance.Frigate docs: Camera setup (detect vs record streams)Used for the detect-on-substream / record-on-main pattern that drives bandwidth and detector load.FS.com: Understanding PoE standards and wattageUsed for the 802.3af/at/bt source-vs-delivered power figures and the switch total-PoE-budget constraint.Frigate docs: Object filtersUsed for the min_score vs threshold distinction, min/max area, and zone/mask tuning to cut false detections.Frigate documentationUsed as the first-party reference for Frigate camera setup, detectors, recording, and configuration.Frigate Docs: Object DetectorsUsed for Frigate detector choices (Coral, Hailo, OpenVINO, TensorRT) and the rule that NVR object detection needs a dedicated accelerator rather than CPU inference.

Diagnostic rule

1. Check the simplest signal first.
2. Change one thing at a time.
3. Prove the stream/reachability before deep config.
4. Stop when the evidence says the risk moved.

FAQ

What's the difference between min_score and threshold?

They're different gates and both matter. `min_score` drops any single detection whose score is below it (filtering weak one-off hits). `threshold` is the score the object's median computed score must reach to be considered a real, tracked object. Raising `min_score` cuts noisy weak detections; tuning `threshold` controls how confident Frigate must be overall. Most guides conflate them.

I added a mask but the false alerts continue — why?

A motion mask only suppresses motion-triggering in that region. Once an object is detected and tracked, the mask no longer hides it. Use masks for perpetual-motion regions (trees, flags, a timestamp overlay), and use object filters (min_score/threshold/area) and zones to control what actually alerts.

How do I stop alerts from areas I don't care about?

Define zones for the areas that matter and require an object to be in-zone before it alerts. Then lean on Frigate's review split — react to 'alerts' (high-value, in-zone) rather than every 'detection'. Use the Debug view with zones enabled to see exactly what Frigate detects and where.