Crash localization and traffic impact assessment via spatio-temporal analysis of connected vehicle data

Abstract

Crashes pose high risk to traffic participants. To enhance rapid response capabilities and improve traffic management, accurate and immediate detection of crashes is essential. This paper investigates a probe vehicle data-based crash detection approach for freeways that solely relies on transmitted position data by a connected vehicle fleet. The algorithm is capable of reconstructing time and location of a crash, even if the vehicles directly involved in the crash are not connected. The method can detect the emergence as well as the resolution of the crash faster than currently used methods. It surpasses traffic information services by detecting crashes 07:40 min earlier and the resolution of the congestion 05:59 min faster. The method also improves spatial uncertainty by detecting precise crash locations instead of incident ranges. The algorithm can classify the detected crashes based on their characteristic traffic patterns, which may facilitate different reaction strategies. Relying solely on global positioning system (GPS) data, it offers a low-cost, real-time solution applicable on a large scale with existing vehicle hardware. A dataset from a connected vehicle fleet, along with ground truth crash data, verifies the proposed method’s results. The algorithm achieves an F1-score of 82.45 % on a dataset containing 1601 congestion patterns with 50 crashes. The paper demonstrates its effectiveness across different regions with varying fleet penetration rates, using empirical examples from freeways in the USA, Germany and the UK.