Driving Style Estimation with Driver Monitoring Systems using Nominal Driver Models

This study investigates the field of Driving Style Estimation (DSE) and its significant implications for enhancing road safety, improving vehicle usability. DSE plays a crucial role for plenty of applications such as the identification of various driving behaviors (distraction, drowsiness, aggressiveness) which are vital for developing driver assistance systems. DSE can play an important role in fields like mission optimization, behavior profiling and virtual coaching also. Due to the high costs associated with obtaining real-world driving data, the research community, in parallel with automotive manufacturers (OEMs), has turned to driving simulators as a cost-effective alternative. These simulators not only provide a controlled environment for collecting driving data but also address sustainability themes, optimizing time and resource allocation in development processes for new algorithms within the automotive industry. This research employs a baseline MATLAB/Python driving simulation framework, enhancing it to develop an Advanced Driver Distraction Warning (ADDW) system through modelling the driver behavior, through estimating additional features related to the interaction between driver-vehicle-environment and combining these features with vehicle data to detect the occurrence of driving distraction. The objective is to evaluate this proposed solution as a complementary approach to traditional camera-based systems, compensating the lack of information about the surrounding environment. The ultimate goal is to provide innovative approaches in compliance with EU regulations for distracted driving, closign detection gaps left by limitations in commercial systems. A physical driving simulator setup facilitates testing with the driver-in-the-loop, enabling high-fidelity data acquisition and processing. Additionally, the development of a Graphical User Interface (GUI) and realistic scenarios utilizing Simulink’s Unreal Engine 4 API contributes to creating a more realistic simulation environment. This immersive setup has enabled the implementation of regulatory-compliant procedures to replicate and acquire distraction episodes, essential for creating machine learning algorithms capable of discerning between appropriate driving conduct and distractions. This research underscores the significance of integration of model-based approaches exploiting additional knowledge that can be considered in machine learning and statistical modeling techniques, thereby advancing the landscape of automotive safety systems.