Automatically evaluating road safety of cyclists using semantic 3D city models

Abstract

This thesis presents a systematic framework for three-dimensional visibility analysis to assess cyclists’ safety at urban intersections using CityGML 3.0 and parking area data extracted using AI-based models trained on the TIAS dataset. Unlike previous visibility studies that are mostly motorist-centric and limited to static or local analyses, this research introduces a scalable, data-driven, and semantically informed methodology that incorporates the geometric and semantic data use of modern 3D city models. The framework overcomes recurring limitations in existing studies by automating visibility quantification and integrating roadside parking data from TIAS into CityGML. The data derived from these datasets represent real-world obstructions with high geometrical accuracy and semantic information. The analysis proposes systematical evaluations of intervisibility between cyclists and drivers across multiple intersections within two study areas in Munich, considering the influence of urban elements such as buildings, vegetation, city furniture, and parked vehicles. Through this approach, the thesis demonstrates how semantic 3D city datasets can enhance reproducibility and scalability in urban visibility studies. The results highlight the applicability of the proposed framework for city-wide assessments, supporting evidence-based and cyclist inclusive urban design and traffic safety planning.