Better Safety Analyses through Smarter Data: Adding Open-Street-View and Traffic-Calibrated Location-Based Services Data to Pedestrian Crash Analysis in Lincoln,NE

Abstract

Pedestrian crashes are a significant concern in the U.S., with pedestrian fatalities increasing and outpacing those of vehicle occupants. This research investigates the potential of new data sources to enhance pedestrian safety analysis and crash modeling. Specifically, it examines the use of StreetLight-calibrated traffic volumes and Mapillary detections of street objects for modeling pedestrian crash counts and severity. By integrating these innovative data sources, the study aims to improve the accuracy and granularity of safety evaluations. Both generalized linear models and machine learning (ML) models, including random forests (RF) and gradient boosting machines, demonstrated acceptable performance and solid portrayal of crash dynamics, with ML models providing better predictive power at the cost of complexity and lower interpretability. Additionally, the weighted RF classifier showed high accuracy in predicting crash severity. Key variables in our analysis encompassed StreetLight volumes and various Mapillary open street-view detections, including traffic signals, crosswalks, advertisement signs, store signs, streetlights, and arrow markings. The association between these variables and crash counts and severity aligns with our understanding of crash patterns. Overall, the research underscores the importance of leveraging detailed, real-world data to improve pedestrian safety analyses and contribute to more effective safety strategies and policy decisions.