A human-centered risk pulse field framework for real-time car-following risk control in connected driving...

Driving risk assessment and control are essential for improving traffic safety in connected driving environments, where vehicles can obtain rich motion information from surrounding traffic participants through vehicle-to-everything communication. However, conventional surrogate safety indicators, such as Time to Collision, mainly rely on simplified kinematic relationships and may fail under zero or small relative-speed conditions. Existing risk field models can describe spatial risk distributions, but they still provide limited representation of driver risk perception and driving style heterogeneity. To address these limitations, this study proposes a human-centered risk pulse field framework for real-time car-following risk assessment and control. First, the formation and evolution of traffic accident risk are analyzed from the perspective of risk pulse energy. Then, an Improved Driving Risk Pulse Field model is developed by integrating potential loss, collision probability, spatial anisotropic correction, acceleration effects, and a perception-modulated risk amplification mechanism derived from relative vehicle motion. A Road Risk Pulse Field model is also established to represent the influence of lane boundaries and road constraints. Based on the field strength of the proposed model, a real-time car-following risk control strategy is further designed by distinguishing safe, warning, and emergency braking states. The proposed framework is evaluated using SUMO simulations under two typical car-following scenarios: periodic speed fluctuations and sudden deceleration of the leading vehicle. The results show that the proposed field-strength indicator provides a more continuous and sensitive risk representation than TTC-related indicators, especially when relative speed is close to zero or acceleration fluctuates sharply. Moreover, the proposed control strategy can regulate the following vehicle's speed in time and reduce collision risk under sudden deceleration conditions. Sensitivity and runtime analyses further indicate that the proposed model has stable risk-response patterns and is suitable for real-time implementation in connected car-following scenarios. The proposed framework provides a more interpretable and human-centered method for driving risk assessment and control.