Abstract
dc:descriptionThe integration of three-dimensional (3D) comprehension and spatial reasoning into intelligent algorithms is essential for addressing various tasks in domains like cultural heritage. Humans possess the cognitive ability to deduce and analyze 3D spatial connections only based on a solitary visual. Nevertheless, the integration of this concept into computer programs does not hold the same validity. By directly manipulating 3D input data, such as 3D point clouds, the reliance on various 2D data in a 3D representation is minimized. 3D point clouds are not only a more suitable depiction of the input data, but they are also becoming ever more accessible. In modern times, laser sensors have been incorporated into consumer gadgets, including mobile phones.The study of dividing and categorizing 3D point clouds is a swiftly developing area because of the growing intricacy of applications and the expanding accessibility of extensive 3D information. Specifically, machine learning and deep learning methods provide significant potential for enhancing the effectiveness and precision of these procedures. Researchers train models using annotated datasets to develop algorithms that can automatically identify and categorize features in point clouds. This process optimizes the conversion of raw data into detailed and functional 3D models.This thesis focuses on using deep learning approaches to classify real-world point clouds that represent medium to large-scale cultural heritage sites. It discusses the accuracy, resolution, and usefulness of various 3D data acquisition technologies for different sorts of objects and situations. This analysis explores the factors that impact the acquisition of 3D data and the methods to enhance them during the process of acquiring 3D data. It outlines the necessary accuracy, precision and level of detail needed for analyzing 3D point clouds using deep learning methods. It enhances the existing datasets by adding new ones that include annotated point clouds. Ultimately, it suggests utilizing a deep learning neural network to do semantic segmentation and classification of 3D point clouds in intricate environments and with diverse items.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chatzistamatis, Stamatis
- Χατζησταμάτης, Σταμάτης
Subjects
dc:subject × 16- 3Δ ψηφιοποίηση
- Νέφη σημείων
- Βαθιά μάθηση
- Ταξινόμηση
- 3D digitization
- Point clouds
- Deep learning
- Classification
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Διεπιστημονικές εφαρμογές επιστήμης ηλεκτρονικών υπολογιστών
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
- Computer Science Interdisciplinary Applications
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57041