Leveraging AI Techniques to Understand and Simulate Driving Behaviours

Understanding and correctly analysing driving behaviours is pivotal for the advancement of autonomous vehicles and the impact assessment of intelligent transportation systems. Indeed, driving behaviour models derived from vehicle trajectories can be employed in several applications, including autonomous driving, traffic management and safety analysis. In this context, artificial intelligence techniques can be used to track vehicle trajectories and have a solid ground truth for analysing and calibrating driving behaviour. Common deep learning architectures can extract vehicle trajectories from camera-based traffic condition sensors. The proposed paper applies AI-enabled tracking of vehicle trajectories and shows how the obtained data can be adopted to identify different families of driving behaviour patterns, which are useful for estimating car-following models and improving the simulation of Intelligent Transportation Systems solutions. The results show that trajectories revealed by artificial intelligence tools, under different traffic conditions, enable the reliable and accurate validation of different modelling approaches.