Aggressive driving detection through Iterative DB-SCAN labeling and supervised pattern recognition

Detection of driver aggressiveness is a significant method which helps to ensure safe driving. Aggressive driving behavior is the cause every year of a vast number of traffic accidents. These traffic accidents which result from this type of conduct are cause of mortality, severe damage and high economical cost. The goal of this thesis is to design an algorithm for aggressive driving detection, capable of recognizing the driver’s behavior. The proposed method is based on sensor features to characterize related driving sessions and to decide whether the session involves aggressive driving behavior. Driving simulators are being increasingly used in recent years by automotive manufacturers and researchers. There are several advantages with using this system, including the increased safety enhanced repeatability, helping researchers significantly reduce time and cost. They play a key role in studies of the driver’s behavior in unstable vehicle conditions and maneuvers. It is exactly this instrument that is employed to collect kinematic data of the vehicle (such as speed, acceleration and heading of the vehicle) in different scenarios. Features are extracted from every observation, which are then labeled by means of unsupervised learning and used to train a pattern recognition neural network. The algorithm is then tested in real-time in the driving simulator.