Research: Development of a Dual-Layer Framework for Conflict-Based Safety Diagnosis and Short-Term Crash Risk Forecasting
This research applies computer vision techniques to process video data and extract road-user trajectories, traffic-flow characteristics, traffic-signal phases and surrogate safety measures, including Post-Encroachment Time (PET), minimum Time-to-Collision (TTC), approaching speed and conflict angle. The identified traffic conflicts are organised within a zone-based intersection framework and combined with road-user type, conflict type, signal phase and traffic exposure to calculate conflict-based risk scores and identify dominant hazardous scenarios. The diagnostic framework then links the observed conflict characteristics to potential crash-risk mechanisms and corresponding infrastructure, operational and signal-control countermeasures using a structured database developed from crash-based studies and road-safety best practices. The diagnostic findings are subsequently used to identify and quantify relevant traffic covariates, such as opposing through–left-turn conflict frequency, short-gap conflict proportion and opposing-through speed, for integration into a Bayesian conditional Extreme Value Theory framework. The aim is to forecast short-term crash risk at the traffic-signal-cycle level by dynamically updating model parameters and quantifying risk through Value at Risk and Conditional Value at Risk. Finally, the diagnostic and forecasting components are integrated into a conflict-based information system that serves as a decision-support tool for intersection risk classification, identification of contributing factors, selection of targeted countermeasures and proactive road-safety management.