Real-time conflict-based Bayesian Tobit models for safety evaluation of signalized intersections

Abstract

The safety of signalized intersections has traditionally been evaluated at an aggregate level by relating historical collision records for several years to the annual traffic volume and the geometric characteristics of the intersection. This is a reactive and macroscopic approach that gives little insight into how important dynamic signal cycle-related variables can affect intersection safety such as the arrival type and the shock wave characteristics. The objective of this study is to develop traffic conflict-based real-time safety models for signalized intersections using several state-of-the-art techniques. Traffic conflicts were measured by multiple indicators including time-to-collision (TTC), modified time-to-collision (MTTC), and deceleration rate to avoid collision (DRAC). Traffic conflict rate was employed as independent variable while traffic volume, queue length, shock wave area, shock wave speed, and platoon ratio of each cycle were used as covariates in the safety models. Four candidate Tobit models were developed and compared under the Bayesian framework: conventional Tobit model, grouped random parameters Tobit (GRP-Tobit) model, random intercept Tobit (RI-Tobit) model, and random parameters Tobit (RP-Tobit) model. The results showed that the GRP-Tobit model performs best with lowest Deviance Information Criteria (DIC), indicating that accounting for the unobserved heterogeneity across sites can significantly improve the model fit. The model estimation results showed that higher conflict rates were associated with various shock wave characteristics (positive sign for shock wave area, shock wave speed, and queue length) and higher traffic volume. Lower conflict rates were related with higher platoon ratio (favorable arrival patterns). The developed models can have potential applications in real-time safety evaluation, real-time optimization of signal control, and connected and autonomous vehicles (CAV) trajectories planning.