Efficient Large Scale Variational 3D Reconstruction (B05)

• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2015 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 251654672

This project investigates real-time 3D reconstruction of large scale scenes. In particular, the aim is the fusion of sparse data representations suitable for high-resolution reconstruction volumes with priors for surface geometry, which typically require a dense representation. While we already have successfully integrated variational methods for this purpose, these methods still employ relatively simple mathematical models for smoothness priors. We now want to focus on machine learning approaches to build priors directly from training data. In particular, we aim at a representation of the local surface structure with deep autoencoders, and exploit their links to the prior probability distribution.

• DFG Programme: CRC/Transregios
• Subproject of: TRR 161: Quantitative Methods for Visual Computing
• Applicant Institution: Universität Stuttgart
• Project Head: Professor Dr. Bastian Goldlücke