A non-intrusive multivariate analysis for monitoring inattentive driving based on physiological signals, driver inputs and visual features: A hybrid deep learning approach

The widespread adoption of autonomous vehicles is inevitable and imminent, but the presence of human supervision in driving will continue to be essential for many years to come. In order to evaluate a driver’s capability to assume control of the vehicle during crucial situations, the monitoring of inattentive driving behavior can be achieved through the adoption of diverse features. This research aims to address the inquiry by developing a non-intrusive multivariate system that can detect driver inattention using hybrid deep learning techniques based on driver inputs established as throttle and brake pedal positions as well as wheel angle, along with the heart rate variability as a physiological signal, and the eye tracking visual feature collected using a driving simulator. The highest performances results have been acquired using the proposed hybrid deep learning architecture on the basis of MLP and CNN. To the authors’ knowledge, there has been a limited interest at assessing inattentive driving behavior based on physiological signals, driver inputs and visual features. As a whole, the proposed system and findings provide new insights into inattentive driving analysis and can be harnessed as a promising tool to improve intervention efforts related to traffic intelligent transportation systems.