How to Fix Frigate Detecting Shadows as People
- Object score threshold (min_score) too low
- min_area too small
- Motion sensitivity too high
Problem Description
Frigate is detecting shadows as people — you get constant false notifications throughout the day, especially at sunrise and sunset when shadows are long. The object detection model misclassifies shadow shapes as person detections, triggering recordings and Home Assistant automations for non-events.
Why This Happens in Real Homes
Frigate detecting shadows as people means the object model is returning a low-confidence "person" on a shadow or lighting change, and your thresholds are permissive enough to accept it. The model isn't perfect on ambiguous shapes, so the fix is tightening the filters so only confident detections count.
Raise the person detection score threshold (min_score/threshold) so borderline shadow detections are rejected, and increase min_area to ignore small or oddly-shaped blobs. Tune motion sensitivity down so shifting light doesn't constantly trigger the detector, and add a motion mask over a problem area that's prone to moving shadows. Improving the detect resolution also helps the model distinguish a real person from a shadow. Higher confidence requirements clear the false positives.
Symptoms
- Shadows detected as people
- False person detections
- Detects shadows/lighting as objects
- Phantom people from shadows
- False positives on shadows
- Detects moving light as a person
- Nuisance person alerts
- Shadow false positives
Recognize these? Here's what usually causes it.
Common Causes
- Object score threshold (min_score) too low
- min_area too small
- Motion sensitivity too high
- Lighting/shadow changes triggering detection
- Model confidence low on shadows
- Detect resolution poor
- No filtering for the false area
- Camera facing changing light
Most fixes happen in the first 3 steps.
Do not globally raise thresholds if only one camera has shadow issues.
Tools & Requirements
Step-by-Step Solution
Add motion masks to shadow-prone areas
Shadows from trees, fences, and buildings move with the sun and trigger Frigate's motion detector, which then feeds frames to the object detector. The object detector can misclassify shadow shapes as people — especially long shadows at sunrise/sunset. In frigate.yml, add motion masks covering areas where shadows regularly appear: cameras: your_camera: motion: mask: - 0,0,200,0,200,300,0,300. Masks are defined as polygon coordinates. Use the Frigate web UI mask editor to draw masks visually instead of calculating coordinates manually.
Increase the minimum object score threshold
Frigate's object detector assigns a confidence score (0.0 to 1.0) to each detection. Shadow-based false detections typically have lower scores (0.3-0.5) compared to real people (0.6+). Increase the minimum score threshold in frigate.yml: objects: filters: person: min_score: 0.6 and threshold: 0.7. The min_score filters individual frame detections, while threshold filters the overall event confidence. Raising both reduces shadow false positives while still catching real people at typical confidence levels.
Set minimum and maximum object size filters
Shadows often create detected 'person' shapes that are abnormally tall and thin, or very small. Set size filters: objects: filters: person: min_area: 5000, max_area: 100000. This filters out detections that are too small (distant shadows) or too large (a shadow covering half the frame). You can also set min_ratio and max_ratio to filter by aspect ratio — real people have a roughly 1:2 to 1:3 width-to-height ratio, while shadows can be 1:5 or wider.
Use zones to restrict detection areas
Instead of detecting objects across the entire camera view: define zones where you actually expect people. In frigate.yml: cameras: your_camera: zones: front_door: coordinates: 100,200,400,200,400,500,100,500. Only trigger notifications and recordings for objects detected within zones. This eliminates false detections from shadows in areas where people never actually walk (far corners, sky area, neighboring property).
Consider a better detection model
The default Frigate model (SSD MobileNet) is fast but has more false positives than larger models. If you have a Google Coral TPU: you are already using the optimized EdgeTPU model, which is generally accurate. If running CPU-only detection: consider adding a Coral USB Accelerator ($30-40) — the EdgeTPU model has fewer shadow false positives than CPU models. For even better accuracy: Frigate 0.13+ supports custom ONNX models and YOLO-based models that are more resistant to shadow detections.
Quick Solutions
Still having issues? This is usually the deeper cause below.
If the sensor still misses events after repositioning, check whether a scheduled 'home' or 'away' mode is overriding the sensitivity setting silently.
False-positive tuning should be time-of-day aware for outdoor cameras.
This issue almost always looks more complex than it is — the majority of cases trace back to a single setting, a stale credential, or a default that shipped wrong.
- Object score threshold (min_score) too low
- min_area too small
- Motion sensitivity too high
- Lighting/shadow changes triggering detection
- Model confidence low on shadows
Before you go — try one of these (they fix most cases).
Official Manufacturer Manual
Frigate provides official product documentation through their online manual rather than downloadable PDF. Access setup guides, troubleshooting steps, and product specifications for your Frigate False Person Detections.
Source: docs.frigate.video
Need More Help? Frigate Support
Note: The contact information below connects you directly to Frigate's official customer support team, not Trunetto. They can help with warranty claims, device replacements, and advanced technical issues.