Microsleep Detector

Project Overview

Microsleep Detector is a mobile safety application designed to prevent driver fatigue incidents. It monitors the smartphone front camera stream to detect eye closure episodes and sudden lapses of consciousness (microsleep).

Computer vision inference runs entirely on-device, enabling continuous operation without requiring active cellular internet during travel.

---

Problems

• High Highway Fatigue Fatalities: Microsleep at highway speeds is a leading cause of severe traffic accidents that drivers often recognize too late.
• Hardware Barrier: Most everyday vehicles lack factory-installed fatigue detection systems.
• Offline Requirement: Mobile connectivity along intercity transit corridors is frequently intermittent, making cloud-dependent computer vision unreliable for safety alerts.

---

Target Users

1. Long-Distance and Night Commuters: Logistics drivers, commercial transport operators, and private travelers driving late hours.
2. Passengers and Families: Providing timely audio warnings to prompt rest breaks before severe drowsiness sets in.

---

Technical Decisions

• On-Device TensorFlow Lite Inference: Quantized MobileNetV2 architecture trained for binary eye state classification (open vs closed).
• High-Throughput Frame Processing: Ingests live front-facing camera frames in Flutter while optimizing battery draw.
• Drowsiness Event Logging: Session dashboard recording eye closure frequencies and duration statistics for driver review.

---

Challenges

• Cabin Lighting Fluctuations: Managing varying ambient illumination, including tunnel transitions and headlight glare at night.
• Latency vs Power Consumption: Maintaining sufficient inference frequency for timely alerts without inducing thermal throttling.

---

Measurable Results

• Complete Offline Functionality: Full real-time eye state classification without network connectivity.
• On-Device Reliability: Continuous frame evaluation with low operational overhead.