3D Model Reconstruction from Indoor Point Clouds via Machine/Deep learning
Suggested elective courses: Laser scanning, Scene unerstanding with UAV or other deep learning courses, programming courses (python)
No fieldwork is included/required.
The research is part of the Digital Twins @ ITC - Ingenuity project. A developer working on the project is available as advisor.
3D models such as Building information models (BIM) are valuable for industry for various planning purposes. Automated methods to reconstruct these from point clouds replace manual modelling work and enable new applications that were not feasible with manual work. In this project, the student is expected to research reconstruction methods to generate 3D models out of point clouds. First step is to run readily available deep learning networks to semantically label the point clouds. Second step is to research a method to generate 3D models based on the labelled points. Generated models are assessed against a ground truth BIM model. See background information: Lehtola, V. V., Nikoohemat, S., & Nüchter, A. (2021). Indoor 3D: Overview on scanning and reconstruction methods. Handbook of Big Geospatial Data, 55-97.
Machine/deep learning methods for indoor point cloud classification. Programming is done with python3. Labelled point clouds and labelled BIM models from the new ITC building (under construction) are made available.