A Driver Fatigue Monitoring and Haptic Jacket-based Warning System

Abstract

Driver fatigue is a major factor in most traffic accidents. This issue has increased the urgency for in-vehicle collision avoidance systems relying on proper driver fatigue detection and warning technologies. Computer vision approaches have been of much interest due to their non-invasive nature for detecting drowsiness. In addition, increased effort has been dedicated to the design of safety systems that warn drivers of various types of collisions. How these systems alert the sleepy drivers when integrated, however, is a crucial component to their effectiveness. A nonintrusive method is proposed in this thesis as a feasible solution to accurately detect fatigue levels and perfectly produce timely warnings. Fatigue progression over time is quantified to more accurate fatigue levels according to reliable PERCLOS measurements in a continuous LBP + SVM based eye state recognition process. Given the quantized fatigue levels, a novel haptic jacket-based alerting scheme is provided to safely convey varying criticality signals. Drivers would have the option to customize haptic jacket settings for the preferred type of feedback perception. This thesis reviews existing approaches, details the proposed system, and finally presents system performance evaluation and usability studies.