Discriminant analysis of driver behavior before safety critical events

This paper presents a novel methodology that can be used to predict safety critical events few seconds in advance in order to warn drivers before their occurrences. Data from the 100-car Naturalistic Driving Study conducted by the Virginia Tech Transportation Institute (VTTI) was used in this study. This Naturalistic database was obtained by instrumenting vehicles and allowing them to be used in normal daily routines, allowing the collection to include normal driving as well as safety critical events. The proposed methodology uses discriminant analysis to accurately distinguish between safe and unsafe driving conditions, using specific combinations of variables that capture changes in driving behavior. A combination of seven variables: longitudinal acceleration, lateral acceleration, vehicle speed, lane offset, yaw angle, range, and range rate, along with their corresponding coefficients were found to create a reliable discriminant score. Two drivers datasets were used in this study. The discriminant analysis resulted in a way to “predict” events as the discriminant scores of the data 6 and 7 seconds, respectively, before a safety critical event show a deviation from normal car following behavior.