A real-time synthesized driving risk quantification model based on driver risk perception-response mechanism

Risk factors within the driver-vehicle–road system are dynamically coupled, with the driver being the most critical factor contributing to system destabilization. However, current traffic risk assessment models struggle to accurately measure the dynamic risk caused by the driver, limiting their applicability in increasingly complex driving environments. Based on the artificial potential field theory, the paper begins its investigation with the driver’s risk perception-response mechanism, and incorporates the effects of risk gain and attenuation to develop a driving behavior dynamic risk quantification model (behavior field). This model is then superimposed with enhanced kinetic and potential fields to construct a real-time synthesized driving risk quantification model under the dynamic coupling of the driver-vehicle–road system, which is validated in various traffic scenarios. The results suggest that: (a) The driving behavior dynamic risk quantification model accurately represents the underlying risks during the driver’s perception, judgment, and decision-making phases. It effectively captures the risk differences between different traffic scenarios and drivers, demonstrating high applicability and sensitivity. (b) The kinetic and potential fields that account for the risk diffusion effect are more consistent with the actual risk distribution characteristics. They can also efficiently represent the risk evolution patterns of influencing factors across diverse scenarios. (c) Compared with the conventional driving safety field and risk evaluation metrics (e.g., steering entropy, jerk, and time to collision), the synthesized driving risk real-time quantification model effectively captures the dynamic coupling of objective traffic environment risks and subjective driving behavior risks on a multidimensional spatiotemporal scale. It provides more robust risk prediction results (R² = 0.988, root mean square error = 0.007). This research can provide a theoretical reference for the automatic analysis of comprehensive traffic risk and the development of more intelligent advanced driver assistance systems.