Thesis award for 3D reconstruction of simplified 3D building models
Obtaining 3D models of buildings is crucial to many urban applications (such as urban planning, solar potential analysis, urban flow simulation), which is still an open problem in remote sensing and the related fields. Existing techniques require high-quality laser scans as input, which imposes challenges to data acquisition. Besides, the building models in the existing platforms (such as Google Earth) are represented by millions of triangles. Such a representation is not friendly for the subsequent processing and application.
Zhaiyu Chen’s MSc thesis aims at obtaining simplified surface models of real-world buildings - models with significantly small numbers of polygonal faces yet can still sufficiently describe the geometry of a building. His thesis proposes a novel framework based on deep learning for reconstructing simplified 3D building models from point clouds, which finds significant applications in the digitalization of the urban environment. This is the first learning-based framework for the 3D reconstruction of urban buildings from point clouds. Though this thesis focuses on urban buildings, the proposed method can already be directly applied to general objects with polygonal shapes. Thus, the research output contributes to not only the field of geoinformation but also computer vision and graphics at large.
Zhaiyu Chen recently received the first prize in the prestigious KNVI/KIVI Thesis Awards for Computing and Information Science.
He received the award for his MSc Geomatic thesis: Learning to Reconstruct Compact Building Models from Point Clouds.
Zhaiyu Chen recently started as a researcher at the chair of 3D Geoinformation, Department of Urbanism. He will start his PhD position at TU Munich (Germany) from February 2022.