National Technical University of Athens (NTUA)
Photogrammetric techniques and algorithms optimization for the development of interactive augmented reality systems
Abstract
dc:descriptionReal-time 3D rendering within immersive Augmented Reality (AR) environments is still a major challenge for computer graphics. New data acquisition and processing techniques of the fields of photogrammetry and remote sensing have drastically increased the requirement for more detailed meshes and the demand for higher display resolutions in recent years. While existing standards deal with specific needs, a comprehensive workflow for integrating such high-fidelity meshes in web-based and Mobile AR (MAR) systems remains an active area of research. This dissertation builds upon existing algorithms introduced in the last few years and develops various novel techniques which significantly improve both visual quality and rendering performance for visualization in the context of AR. The advances can be summarized as three main contributions. Firstly, batched multi-resolution model structure and view-depending loading are introduced as a lightweight 3D format and a high-performance rendering method for web-based AR applications respectively. The proposed method handles efficiently high-resolution and large-sized meshes that have been converted into the dedicated data type, supporting both natural feature tracking and location-based AR. The solution prioritizes the highest possible visual fidelity, while simultaneously delivering interactive frame rates and effectively managing the resource constraints inherent in mobile platforms and network bandwidth. The second contribution lies in combining traditional computer vision methodology with either low-level data organization, management and synchronization techniques or modern Deep Learning (DL) to enable the use of photogrammetric meshes as 3D overlays and optimize pose estimation. Two 3D engines are developed: a cross-platform Vulkan-based graphics engine with multithreading and efficient data transfer mechanisms, and an OpenGL ES-based engine for low-memory, limited CPU/GPU environments, employing data organization, instancing and texture compression. The OpenGL ES-based engine is extended as the basis of a lightweight pattern-based MAR prototype. The methodology tests and compares widely used feature detectors and binary descriptors combinations for efficient real-time recognition. Furthermore, a MAR prototype that combines visual Simultaneous Localization and Mapping (vSLAM) algorithm with DL-based object detection improves tracking performance in challenging outdoor environments, featuring comprehensive evaluation against other DL models. Thirdly, the proposed technical specifications, best practices and guidelines for the creation and management of such meshes in AR applications aim to pave the way for the integration of high-fidelity photogrammetric data on immersive technologies and evolve the sector of photogrammetry and 3D visualization.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Boutsi, Argyro-Maria
- Μπούτση, Αργυρώ-Μαρία
Subjects
dc:subject × 15- Επαυξημένη πραγματικότητα
- Φωτογραμμετρία
- Τρισδιάστατα ( 3Δ ) δεδομένα
- Οπτικοποίηση
- Augmented reality
- Photogrammetry
- 3D models
- Rendering techniques
- Visualization
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Τεχνολογία μέσων
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Media Technology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/59886