Realistic traffic predictions

The safety of road traffic is significantly influenced by advancing automation and multimodal traffic concepts. Fraunhofer EMI is developing algorithms to precisely analyze critical traffic situations of the future.

Automation and multimodal concepts are revolutionizing road traffic. This makes the development and testing of new functions increasingly challenging.

Fraunhofer EMI therefore relies on agent-based simulations that reproduce realistic traffic situations. The focus here is on the detailed simulation of critical traffic scenarios.

How to generate more realistic traffic simulations? Agent-based models are often used for microscopic traffic flow simulation, in which vehicles are divided into agent classes such as cars or trucks. Each vehicle is assigned behavioral parameters that are drawn from statistical distributions of the respective agent class. The parameter values of these distributions must be defined before the simulation. EMI has developed a method to determine optimal parameter values of the behavioural models through numerical optimization based on traffic data. The aim is a realistic simulation that maps statistical traffic variables.

How to recognize critical situations in road traffic? A key feature of realistic traffic flow simulation is the ability to depict critical situations using the integrated behavior models - from potentially dangerous situations to actual accidents - in a statistically comparable way to real traffic. In order to take these aspects into account in the simulation, the behavior models used must be able to depict critical situations. It is also necessary to have sufficient data on critical situations in road traffic. The analysis and evaluation of the criticality of driving scenarios in existing and future data therefore represent central challenges in the implementation of realistic simulations. Fraunhofer EMI is tackling these challenges and developing evaluation algorithms for the automated identification of critical driving scenarios in traffic environments of varying complexity.

How can behavioral models improve the safety of pedestrians and cyclists in traffic simulations? Pedestrians and cyclists are among the vulnerable road users who are insufficiently considered in current traffic simulations. Fraunhofer EMI is developing behavioral models for this group based on simulations of the flow of people at major events. AI methods such as reinforcement learning are used to teach virtual pedestrians how to cross roads safely.

How to better analyze and manage crowd flow during major events? Fraunhofer EMI accompanied this year’s Juicy Beats Festival, working alongside Fraunhofer IOSB to create a 3D reconstruction of the site. Utilizing this model, they conducted visibility analyses and simulated crowd flow based on video recordings of visitors and influx data provided by the event organizers.