Driver Behavior Classification under Cut-in Scenarios Using Support Vector Machine Based on Naturalistic Driving Data 2019-01-0136
Under Cut-in scenarios, vehicle safety is affected by both longitudinal relative motion and lateral relative distance with the target vehicle (namely the cutting-in vehicle). Human driver can detect the cut-in intention of target vehicle and then take appropriate maneuvers to reduce potential collision risk. In some dangerous Cut-in situations, driver may brake very hard. While driver may take no brake action in normal Cut-in situations. This paper proposed a method to figure out under what circumstance the Cut-in case is dangerous condition or normal one. 78 dangerous cases and 249 normal driving cases were extracted from naturalistic driving database. In each dangerous case, the largest deceleration of subject vehicle is over 0.4g; and in normal driving cases, subject vehicle may keep accelerating and the largest decelerations of some brake cases are less than 0.3g. It was found that in dangerous cases, drivers are most likely to brake when target vehicle right cross lane line; while in normal driving cases, the brake time point is mainly when target vehicle right crosses lane line or when target vehicle intrudes into 1/4 of the lane of subject vehicle. And longitudinal safety is related to THW (Time Headway) and relative velocity. Support Vector Machine (SVM) was utilized to divide danger zone and normal zone on THW-relative velocity plane. And on the normal zone, boundary between brake zone and non-brake zone was set according to driver behavior. The zone division in this paper can provide theory evidence for Advanced Driver Assistance Systems (ADAS) control strategy to make the systems tackle Cut-in in a more human-like way. Besides, the SVM model proposed can also facilitate evaluation of automated vehicle.