From Real-World Traffic Data to Scenarios in the Context of Automated Vehicles

Kurzfassung

The large-scale introduction of automated vehicles (AVs) on public roads is an ambitious and challenging goal. This technology aims to significantly contribute to increased traffic safety and comfort, while also serving as the foundation for further innovative mobility concepts. But, one of the biggest challenges in introducing such systems is ensuring their compliance with regulatory requirements, since it is essential that these systems function correctly and safe. This is particularly demanding for higher-level automated vehicles, which must navigate public roads safely without a human fallback option. Within the framework of homologation of automated vehicles, an approach has been established where the driving function or the automated vehicle is tested in specific scenarios. However, the availability of an extensive dataset with diverse scenarios is a critical prerequisite. These scenarios can be defined based on different data sources. One method for collecting and evaluating scenarios involves using real traffic data. Such data is highly valuable as it realistically reflects the behavior of human road users and also includes atypical behavior or even critical conflicts between participants. For collecting real traffic data various methods are available, each with different strengths and weaknesses. One method is infrastructure-based traffic data collection using road side units, which main advantage is capturing traffic events comprehensively and over an extended period. This allows for the continuous and simultaneous capture of multiple road users and their interactions. Consequently, this method enables a detailed description of scenarios and the identification of rare phenomena, which are particularly important for validating driving functions. However, this continuous stream of traffic data must be systematically processed to create a comprehensive collection of scenarios. This thesis presents a methodology for representing traffic data collected in the real world based on scenarios and their systematic identification from real-world traffic data. A hierarchical data model is used for this purpose, which semantically describes traffic data at four different levels of abstraction. Various approaches to defining and identifying phenomena at those abstraction levels using real traffic data are presented. Furthermore, a modular platform is introduced that integrates these different methods to continuously identify and analyze scenarios in real traffic data and various environments. The procedures and methods presented in this work are individually evaluated using real-world problems. Their integration through the modular platform demonstrates the suitability of the proposed approaches for identifying and analyzing scenarios in different environments and for various research questions. Overall, the results show that the proposed methodology enable the systematic identification and representation of scenarios from real-world traffic data contributing to building a large-scale knowledge base of scenarios.