Driver drowsiness, compounded by the high workloads and stress of the ever-increasing complexity of car and traffic environments, is a major cause of severe accidents. The objective of the project described in this paper is to develop reproducible and flexible methods for studying the relationships between physiological driver states and human-factor issues in a driving environment. For reasons of safety and reproducibility, a laboratory-based driving simulator is being used for the project experiments. Initial experiments were conducted with a cohort of about 60 healthy male subjects aged 22 to 28 under carefully controlled conditions. Performance was measured before, during, and after a 120 km stretch of stimulus-deprived, foggy highway that was intended to induce fatigue and stress. Across all trials 69% of the subjects experienced sleep events lasting several seconds, and 7 potentially fatal crashes occurred. Lane tracking behaviour degraded by a factor of 2 to 3 prior to each crash. Much of the extensive data acquired by these experiments remains to be analysed using both standard statistical techniques and high-dimensional clustering algorithms. ALISA image-processing software is being applied to video images of the driver eyes and face to detect the onset of sleep and other critical situations.
Abstract