Pre-Intrusion Detection: Beyond Simple Motion Alerts
Conventional CCTV systems trigger on any movement, generating hundreds of false positives daily. Behavioral Pattern Recognition shifts focus from object presence to behavioral context — identifying pre-intrusion indicators before a break-in occurs.
Frigate NVR: The Foundation
Frigate is an open-source NVR running real-time YOLO object detection on Coral TPU. Its Zone System enables sub-regions within camera views for targeted behavior analysis.
Pre-Intrusion Indicator 1: Zone Loitering Duration
A person remaining in the sidewalk zone outside a home for >30 seconds without passing through is statistically anomalous. Home Assistant automation tracks zone occupancy duration: if the person sensor remains active in after a 30-second delay, a pre-intrusion alert fires with a Frigate thumbnail snapshot.
Pre-Intrusion Indicator 2: Vehicle Parking Anomaly
Unknown vehicles parking outside during unusual hours (23:00–05:00) or for extended durations (>15 minutes) trigger investigation. Integrate CodeProject.AI or Rekor Scout for local Automatic License Plate Recognition (ALPR) to compare against a known vehicle whitelist. Unknown plates parking during anomalous hours contribute to the risk score.
Pre-Intrusion Indicator 3: Exterior Lighting Anomaly
Flashlight use near perimeter walls, fences, or windows at night is a classic pre-intrusion behavior. Detect with: - BH1750 I2C lux sensor (150 THB) measuring sudden brightness spikes in dark hours - Frigate motion detection sensitivity tuned for nighttime light flash patterns - Home Assistant History Stats flagging 3+ motion bursts within 5 minutes at night
Pre-Intrusion Risk Score
A cumulative scoring model aggregates signals: loitering >30s (+30 pts), unknown vehicle parking >10min (+25 pts), exterior light flash at night (+20 pts), multiple motion bursts in 5 minutes (+25 pts). At ≥50 points: LINE notification with snapshot. At ≥75: 60-second video clip + critical alert. At ≥100: all perimeter lights on + full recording.
Real-World Results
In a Lat Phrao district pilot, the system detected suspicious activity an average of 8–15 minutes before any incident across all 4 cases, while reducing false positives by 85% compared to simple motion detection.