Digital twins and virtual replicas of urban environments have a broad range of applications across various
industries, including crisis management, evacuation planning, and emergency response training. Their ability to
integrate physical environments with digital representations enables organizations to simulate, monitor, andoptimize real-world assets and operations. However, creating detailed and functional digital replicas of outdoor
environments, preferable in LoD2 which includes detailed roof shapes, often requires extensive manual effort,
particularly when open data sources lack essential metadata including structural dimensions or semantic
features of buildings.
To address this challenge, leveraging Digital Surface Models (DSM) in combination with building footprints
offers a promising solution to recreate a highly detailed representation of the urban environment. These data
sources are often commissioned by the respective city authorities and publicly available. However, classical
rule-based approaches such as plane fitting followed by roof primitive matching are prone to errors and
unreliable when data is noisy.
Your responsibilities:
Therefore, the objective of this work is to investigate which methods can be employed to automatically infer the
roof type of buildings from a DSM to support the scalable creation of high-fidelity digital twins for safety-critical
applications:
• Evaluating current advances in inferring the roof type based on ordered/unordered point clouds and triangulated meshes (both can be derived from the DSM)
• Developing a pipeline to infer the roof type based on DSM and building footprints
• Comparing proposed pipeline with a classical rule-based approach
• Assessing the practical effectiveness of the proposed pipeline in representative use cases
Required Qualifications:
• Currently enrolled as a student in computer science, mathematics, or similar major
• Basic knowledge of common 3D representation formats in computer graphics, such as meshes, point clouds, and voxels
• Basic knowledge in Deep Learning and Computer Vision
• Basic Python skills (ideally with respect to deep learning applications, e.g., PyTorch or Tensorflow)
We look forward to getting to know you!
If you have any questions about this position (Vacancy-ID 4677) please contact:
Tobias Koch
Tel.: +49 2241 20148 55
