A large-vehicle-coupled driving risk field model for risk assessment at signalized intersections

OBJECTIVE: The specific maneuvering characteristics and visual occlusion of large vehicles significantly increase driving risks at signalized intersections. However, existing risk assessment models generally fail to capture the distinct risks posed by such vehicles, especially regarding visual obstruction and blind zones. METHODS: Based on safety potential field theory, a large-vehicle-coupled driving risk field (LVC-DRF) model for signalized intersections is proposed that integrates visual occlusion and blind-area constraints with vehicle specific kinematic parameters. Based on this model, a Relative Driving Safety Index (RDSI) is proposed for continuous risk assessment across all traffic scenarios, avoiding the discontinuity and environmental insensitivity found in conventional surrogate safety measures. The model is calibrated using real trajectory data, and its performance is evaluated against ETTCI through empirical tests and ablation studies. RESULTS: Results show that RDSI detects more dangerous samples than ETTCI, matches 95.6% of high-risk events identified by ETTCI, and issues warnings 0.84s earlier on average. Ablation analysis shows that both vehicle-specific kinematic parameter modeling and the large-vehicle constraint field contribute to the representation of large-vehicle-related risk. Incorporating the constraint field increases the overall proportion of dangerous samples from 17.0% to 18.8%. CONCLUSIONS: The LVC-DRF model comprehensively accounts for vehicle interactions, environmental constraints, and large-vehicle effects in signalized intersection risk assessment, and offers advantages in both temporal and spatial continuity. In addition, RDSI exhibits strong consistency with ETTCI in detecting high-risk events and provides earlier warnings for events identified by both indicators. This work contributes to improving the accuracy of risk assessment for vehicles in complex mixed traffic flows and provides both theoretical and data-driven support for proactive safety prevention and control.