Applying evolutionary methods for early prediction of sleep onset

Driving safety can be achieved by predicting e.g., about 40,000 per year in the USA alone [1], but such imminent falling asleep at the wheel. Several methods of statistics underestimate the number because they do not include most crashes due to fatigue or those where early detection have been investigated by continuous monitoring of physiological and behavioral parameters. drowsiness was not apparent, so this kind of risk actually Requirements for noninvasive, unattended, personal adap- involves millions of drivers. If fatigue and drowsiness are detected early, accidents can be prevented by implement- tation need to be met, along with the effectiveness of the ing countermeasures based on onboard safety devices, such detection method, in order to perform reliably when as automatic cruise control, or simply on acoustic alarms if applied. Because wakefulness and sleep are reflected in the system is not integrated into the onboard cruise elec- several human physiological conditions, such as cardiac tronics. To detect sleep onset early, the driver’s state needs activity, breathing, movement, and galvanic skin conduc- to be continuously monitored. tance, captured bioelectric signal features were extracted. Hard computing methods were applied to develop early A fuzzy decision-fusion logic was tuned to make infer- at-the-wheel-sleep detection, mostly based on behavioral ences about oncoming driver fatigue and drowsiness. The (head, eye, and hands movement) [ 2 ] and/or physiological evolving fuzzy neural network paradigm was applied to the (heart-rate variability and breathing-rate variability) [ 3 – 5 ] previous developed framework to improve reliability while measurements and by applying signal-processing algo- keeping target system complexity low.