DriverSonar: Fine-Grained Dangerous Driving Detection Using Active Sonar
summary
System uses an off‑the‑shelf smartphone speaker and microphone to emit 1 ms chirp pulses with ≥5 ms intervals and track head and hand echoes within ~1 m, separates echoes via relative correlation coefficient processing, extracts time‑domain motion features and matches them with DTW. Prototype runs in real driving and simulator with 30 volunteers over 12 months, recognizes nodding, yawning and abrupt steering adjustments, and reports up to 93.2% precision with 3.6% false acceptance.
AI-written summary · about