Modeling conflict risk with real-time traffic data for road safety assessment: a copula-based joint approach

Abstract

This study proposes a conflict-based traffic safety assessment method by associating conflict frequency and severity with short-term traffic characteristics. Instead of analysing historical crash data, this study employs microscopic trajectory data to quantify the relationship between conflict risk and traffic characteristics. The time-to-collision (TTC) index is used to detect conflicts, and a severity index (SI) is proposed on the basis of time-integrated TTC. With SI, the k-means algorithm is applied to classify the conflict severity level. Then the severity of regional conflict risk is split to three levels. Zero truncated Poisson regression and ordered logit regression methods are employed to estimate the effects of short-term traffic characteristics on conflict frequency and severity, respectively. Furthermore, the copula-based joint modelling method is applied to explore the potential non-linear dependency of conflict risk outcomes. A total of 18 copula models are tested to select the optimal ones. The HighD dataset from Germany is utilized to examine the proposed framework. Both between-lane and within-lane factors are considered. Results show that the correlations between traffic characteristics and conflict risk are significant, and the dependency of conflict outcomes varies among different severity levels. The difference of speed variation between lanes significantly influences the conflict frequency and severity simultaneously. Findings indicate that the proposed method is practicable to assess real-time traffic safety within a specific region by using short-term (30-second time interval) traffic characteristics. This study also contributes to develop targeted proactive safety strategies by evaluating road safety based on conflict risk, and considering different severity levels.