DBALapisco: An Abnormal Driving Detection System for Safety Validation of Autonomous Vehicles

Driver behavior analysis is an important topic for both operation and safety validation of autonomous vehicles. The main causes of traffic accidents are drunk, fatigued and distracted drivers, which we call abnormal driving. We aim to detect abnormal driving for filtering most safety-relevant scenarios. To do so, we use the HighD, InD, ExiD and RounD datasets, which contain vehicle trajectories from German road segments. We process the data to represent abnormal driving and train the MLP neural network. After all, we launch the DBALapisco, a system ready to be used as plugin for autonomous vehicles to detect abnormal driving and by engineers to develop routine analyzes.