Gomes, Pedro de Medeiros;
(2025)
Graph-based Learning for Dynamic Point Clouds.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
Gomes_10208890_thesis_revised.pdf Download (18MB) | Preview |
Abstract
Autonomous systems, ranging from self-driving cars to household robots, are promised to play a critical role in our future society. These systems are expected to understand their surrounding environments and automatically adapt to changes. This means the ability to learn the scene behaviour is essential. A practical way for autonomous systems to capture the surrounding environment is by representing the 3D space as a point cloud. A point cloud is a set of independent points in 3D space. While this point independence makes the point cloud a very flexible representation of 3D objects, it also makes the point cloud very challenging to process, since there are no explicit relationships between points. The goal of this thesis is to develop new tools and techniques that enable autonomous systems to learn the dynamic behaviour of a 3D scene. To this end, we tackle the challenges associated with dynamic point cloud processing. More specifically, we propose a graph-based approach to process the irregular point cloud data that we can leverage to learn relationships between the points. We apply our approach to dynamic point clouds of i) high-resolution human bodies acquired from LiDAR scanners and ii) low-resolution noisy indoor scenes acquired from millimetre-wave radars. For the human body point clouds, we design a framework for motion prediction based on hierarchical spatio-temporal graphs and adaptive feature combination to control the composition of local and global motion. For the indoor scenes point clouds, we introduce a framework able to classify a point as a true object or a false point generated by additional reflections, based on its dynamic behaviour. Overall, this thesis demonstrates the advantages of graph-based learning for dynamic point cloud processing.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Graph-based Learning for Dynamic Point Clouds |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Copyright © The Author 2025. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Electronic and Electrical Eng |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10208890 |
Archive Staff Only
 |
View Item |
