Simulation of Urban Crash Occurrence Based on Real-World Crash Data

Abstract

The intelligent application of simulation is of central importance for the successful development and testing of automated driving functions. Realistic virtual environments are required to assess and optimize both the efficiency and safety of automated driving functions in real-world traffic situations. While existing traffic flow simulation frameworks excel at evaluating traffic efficiency, the implementation of human failure models and traffic safety aspects is a current field of research. In this publication, the occurrence of human failures is inferred from real-world crash statistics and introduced into traffic simulation. A realistic traffic simulation setup of the city of Ingolstadt, Germany, is used as a basis for this simulation of crash occurrence. Focusing on intersections as the most important urban crash hot spots, the relation between human failures and the occurrence of collisions is estimated for each conflict point in the simulation network. From crash statistics, the distributions of crash quantities and types across the intersections in the simulation network are calculated. An Iterative Proportional Fitting algorithm is used to project crash counts available at the intersection level onto the “conflict level,” determined by intersecting traffic streams within intersections. Human failures are generated and applied to traffic participants in the simulation using a Monte Carlo selection. The results demonstrate the functionality of the method for calibrating models for realistic crash occurrence in traffic simulation. This methodology provides a basis for simultaneous evaluation of both traffic efficiency and traffic safety impacts of future developments in urban traffic networks.