A SMART RASPBERRY PI-BASED SYSTEM FOR MICROSLEEP DETECTION USING EYE-TRACKING AND IMAGE PROCESSING

Abstract

Microsleep is a brief, involuntary sleep episode that often occurs without the individual's awareness, particularly in monotonous situations such as long-distance driving. This condition poses a significant safety risk as it can result in a sudden loss of vehicle control. This project develops a microsleep detection system based on the Raspberry Pi 4, designed to detect early signs of microsleep and deliver immediate alerts to the user. The system integrates camera-based eye-tracking technology with facial recognition and advanced image processing algorithms to monitor physiological indicators such as eye closure, blink rate, and head movement. Visual data from the camera is utilized to identify slow blink patterns and the absence of visual responses that are consistent with microsleep symptoms. When an episode is detected, the system activates a dual-mode alert mechanism using visual signals (LED) and auditory warnings (buzzer) to provide real-time feedback and help prevent potential accidents. The system design balances user-friendliness with physical durability and functional efficiency by integrating a webcam, LEDs, buzzer, and a custom protective casing. The software algorithms are also optimized to achieve high sensitivity and accuracy during the detection process. Research findings indicate that the developed microsleep detection system shows strong potential as an effective and affordable safety tool, particularly in transportation and industrial sectors, with the ability to reduce incidents caused by human error. Tested in a controlled environment with participants engaged in continuous visual tasks, the system achieved 87% accuracy, 90% sensitivity, and 84% positive precision, while maintaining a lower false positive rate of 11% compared to camera-only systems. With an average response time of 1.4 seconds after detecting eye closure beyond 2 seconds, and by combining Eye Aspect Ratio (EAR) readings with reflective IR sensor data, the system demonstrated enhanced robustness and reduced reliance on a single signal source. Further research is recommended to integrate advanced sensors, refine algorithms, and develop intelligent feedback mechanisms to improve real-world performance.