Assessing Roadside Safety With Computer Vision: FHWA Ratings as the Key Predictor of Rural Road Departure Crashes and Severity

Abstract

This is the first study to evaluate the effectiveness of the Federal Highway Administration (FHWA) roadside safety rating system in predicting Road Departure (RD) crashes on rural roads. The research employs a two-step framework: first, a computer vision model was used to extract detailed information on clear zones, rigid obstacles, side slopes, and safety barriers from roadway images. Next, the extracted data was integrated with crash records for statistical analysis. The FHWA safety rating system, which combines these features, shows a significant correlation with rural RD crash frequency and severe injury rates, as confirmed by Spearman correlation coefficients. Furthermore, using the negative binomial regression model, the safety rating emerged as the strongest predictor of rural RD crashes and their severity compared to individual roadside features, underscoring its value in assessing crash risk. With its seven categories, the FHWA safety rating system provides a more comprehensive predictor of rural RD crash risk, making it an essential tool for identifying high-risk locations and prioritizing safety interventions.