Real-time risk estimation for active road safety: Leveraging Waymo AV sensor data with hierarchical Bayesian extreme...

Area of Science:

• Traffic Safety Engineering
• Statistical Modeling
• Autonomous Systems

Background:

• Traditional traffic risk assessment relies on historical crash data, which may not capture real-time risks effectively.
• Existing methods often lack integration of diverse roadway geometries, crash patterns, and 2D vehicle dynamics.
• Autonomous vehicle sensor data offers a high-fidelity source for near-miss event detection.

Purpose of the Study:

• To develop and validate a real-time traffic risk estimation framework using Extreme Value Theory (EVT).
• To integrate 2D vehicle dynamics and roadway characteristics into near-miss risk assessment.
• To improve the accuracy and generalizability of traffic safety analysis for autonomous and conventional vehicles.

Main Methods:

• Utilized a 2D time-to-collision (TTC) near-miss indicator from the Waymo motion dataset.
• Applied univariate Generalized Extreme Value (UGEV) distribution models with Block Maxima (BM) sampling.
• Incorporated vehicle dynamics as covariates in non-stationary hierarchical Bayesian structure with random parameters (HBSRP) UGEV models.

Main Results:

• The HBSRP-UGEV models demonstrated superior performance compared to other methods.
• A 6.43-10.56% decrease in Deviance Information Criterion (DIC) was observed, particularly for short-duration traffic segments.
• The inclusion of dynamic vehicle behaviors and random effects significantly enhanced real-time traffic risk estimation.

Conclusions:

• The developed generalized real-time EVT model offers a precise and adaptable tool for network-level traffic safety analysis.
• This approach bridges the gap between active and passive safety measures in intelligent transportation systems.
• The framework provides a robust method for estimating traffic risks by accounting for vehicle heterogeneity and dynamic behaviors.