Study On Vehicle Collision Predicting Using Driver Behavior Characteristics

The combination of activity and passive technology has become an attractive field in vehicle safety recently, which adjusts the restraint system before collision to reduce injury based on the prediction of collision. The active safety sy stem which based on the detection of driving environment is too expensive to widely used in vehicle due to its complex construct. Studies show the time between the occurrence of collision reason and collision is up to few seconds, and most of drivers brake and turn before collision. The injury of pedestrian collision and side to side crash have a great proportion in the statistics of traffic accidents, and collisions caused by driver whose driving experience is less than 1 year are more than other phase of driving experience. Therefore this paper plan to develop an economic collision prediction system which based on the driving behavior in pedestrian collision and side to side collision in order to applied in vehicle widely.In this paper, virtual driving scene of normal and emergency(about 15 km) which contain 17 intersections had been build thorough in UC-win/road based on actual driving environment. Both pedestrian collision and side to side collision in emergency scene have 4 different times to collision(1.5 s, 2.5 s, 3.0 s and 3.5 s) and 2 directions of obstacles. Braking acceleration, angle velocity of brake pedal, angle of steering wheel and angle velocity of steering velocity were defined as driving behavior, box plot and Mann-Whitney rank-sum test were used to analyze the effect of gender and driving styles on driving behavior. Correlation analyses had been applied to discuss relationship between driving behaviors. Braking acceleration, angle velocity of brake pedal, angle of steering wheel and angle velocity of steering velocity were defined as input parameters in RBF neural network to identify emergency situation.The result showed that gender have effect on the distribution of the mean value of the angle of steering wheel in emergency scene only, b oth anxious and cautious driving styles had influence on the distribution of the mean value of the angle of brake pedal in emergency scene only. The difference of distribution of all driving behaviors between normal and emergency driving scene was signific ant. Moderate correlation was found in braking acceleration and angle velocity of brake pedal, as well in angle of steering wheel and angle velocity of steering velocity. The RBF neural network build in this paper showed an accuracy of 98.1% in driving sce ne recognition. The recognition of driving scene was controlled by the angle velocity of brake pedal mainly; gender and driving style had tiny influence on the recognition.Gender and driving style had tiny influence on driving behavior the rec ognition of driving scene. Collision prediction system that based on driving behavior could predict collision accurately, and had great value for popularization.