Online risk estimation of critical and non-critical interactions between right-turning motorists and crossing cyclists by a decision tree. (December 2021)

• Record Type: Journal Article
• Authors: Saul, Hagen; Junghans, Marek; Dotzauer, Mandy; Gimm, Kay
• Is Part Of: Accident analysis and prevention. Volume 163(2021)
• Publication Date: 2021-12
• Subjects: Risk estimation -- Right-turning vehicles and crossing cyclists -- Vulnerable road user -- Surrogate measures of safety -- Decision tree; Accidents -- Prevention -- Periodicals; Accident Prevention -- Periodicals; Accidents -- Prévention -- Périodiques; 363.106
• DOI: 10.1016/j.aap.2021.106449
• Languages: English
• ISSNs: 0001-4575

Abstract

Highlights: Development of a real-time and infrastructure-based early warning system for both right-turning vehicles and crossing cyclists that allows estimating risks. Application of a data-driven and machine learning approach encoding human risk perception in a decision tree. Identification of the most supporting features for the prediction of conflicts: Gap time (i.e. predicted post encroachment time), velocity, and distance from conflict point. Evaluation of the warning system in test drives showed reliable results. Abstract: One of the most critical situations in urban areas is when motorists turn right in an intersection and cyclists cross the road. Many of those crashes result in severe consequences for cyclists. In order to increase the safety of cyclists, especially in the case of conflicts with right-turning vehicles, an online infrastructure-based assistance system may be a promising solution warning drivers and cyclists when a conflict or crash is predicted. By means of automated video traffic detection, the resulting trajectories of road users can be analysed and a warning can be sent to vehicles and cyclists equipped with vehicle-to-anything communication (V2X) when a high risk is estimated. An approach for online risk estimation was developed combining the surrogate measure of safety (SMoS) gap time (GT) with trajectory prediction-based estimates of the time-to-arrival (TTA) or distance to conflict point (DCp) and velocity (v). A decision tree as classifier of risk levels based on the previous named risk features was trained to model the risks perceived by humans. Expert ratings of traffic conflict scenes were used to build a model, apply the model, and improve it in the field. The warning system was evaluated by test drives in real traffic at the urban AIM Research Intersection in Braunschweig, Germany. In general, the system warned reliably. In approximately 67% of the trials, it was assessed as helpful.