SenseGuard

Phone & wearable sensing • On‑device anomaly detection • Fall/crash alerts

Sensors

Motion sensors:
unknown
Orientation:
unknown
Geolocation:
unknown
Battery level:
n/a
Watch via Bluetooth:
not connected
Real sensors: DeviceMotion/Orientation, Geolocation. Simulator fills in if unavailable or for demos.

Live stream

Accel magnitude:
0 m/s²
Gyro magnitude:
0 °/s
Jerk (acc change):
0 m/s³
Speed estimate:
0 m/s
Position:
n/a

Model training

Record a short window (30–60 s) of normal activity. We’ll compute a feature set and learn a distance-based profile. Threshold is chosen from the 99th percentile on normal data.
Samples recorded:
0
Windows:
0
Status:
idle
Window: 1.0 s, hop 0.25 s
Features: mean, std, max, SMA, jerk, gyro, posture
Detector: kNN distance (k=5)

Anomaly detection

Anomaly score:
0.00
Threshold (99th pct):
n/a
State:
not running
Debounce: 1.2 s above threshold to trigger. Cooldown: 15 s. Fall pattern check: high impact + posture change + inactivity.

Simulation playground

Simulated signals mirror realistic amplitudes: impact spike in acceleration, abrupt orientation change, then stillness.

Alerts and automations

This demo composes messages with deep-links (SMS, WhatsApp, email) and can call a webhook. Actual auto-sending depends on platform capabilities and user interaction.

Use cases

  • Elderly fall: High-g impact (≥ 2.5g), device orientation change (portrait → flat), then inactivity ≥ 10 s.
  • Bicycle crash: High jerk, erratic gyro oscillations, speed drop, optional GPS off-route.
  • Hard brake: Forward decel spike, transient orientation pitch, quick recovery.
The model scores anomalies; pattern rules refine to specific events and reduce false positives.

Privacy & on-device

Data stays in your browser. No cloud training by default. Use the optional webhook to integrate with your system. Revoke permissions anytime in your browser settings.

Event log