A framework for driver specific inference of danger at signalized intersections

Short range communication network systems and their application to in-vehicle warning devices represent an emerging technology that is indispensable for future improvements in safety to motorists. It is widely reported that consumer acceptance of such devices is greatly influenced by behavioral reactions to the warning. The acceptable level of intrusiveness or annoyance, and tolerance for false alarms, may be quite different for different people and will depend greatly on individual drivers and their behavior pattern. The objective of this paper is to incorporate the human element in the protocol methods of future warning devices. We develop a statistical framework to classify human behavior based on prior driving patterns and provide a model on how humans infer danger from multisensory information. The focus is on "in-vehicle crash avoidance warnings" that are triggered by the likelihood of vehicles running through red lights at signalized intersections.