Muscoi, Vlad-Nicolae (2025) Generating synthetic images featuring cracks in masonry surfaces using Unreal Engine 5. Master's Internship Report, Computing Science.
|
Text
CrackGenerationwithUnrealEngine-1.pdf Download (21MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming Muscoi.pdf Restricted to Registered users only Download (185kB) |
Abstract
Masonry structures constitute a significant portion of our architectural heritage. Timely crack detection is essential to preserving their structural integrity. Current inspection methods rely predominantly on manual assessments, which are time-consuming, expensive, and subjective. Recent advances in deep convolutional neural networks (CNNs) have significantly improved automated crack detection. However, the principal limitation remains the availability of annotated data. To address this, a recent study has employed Blender’s 3D engine to procedurally generate synthetic masonry datasets. While augmenting up to approximately 30% of a training set with synthetic images yields a small drop in F1 score, further increases lead to a marked decline in overall performance. This degradation is likely due to limited scene diversity in the synthetic data. In this study, we introduce a novel framework built in Unreal Engine 5 to enhance the building of virtual scenes for use in generating synthetic masonry datasets. By leveraging the engine’s advanced features—including Nanite for high-fidelity displacement, Blueprints and Materials for procedural crack generation, and the Procedural Content Generation (PCG) system for automated placement of scene clutter—we aim to improve both the realism and variability of synthetic datasets. This framework offers a scalable and flexible approach to producing training data for deep learning-based crack detection models.
| Item Type: | Thesis (Master's Internship Report) |
|---|---|
| Supervisor name: | Kosinka, J. and Boerema, D.H. and Bal, I. |
| Degree programme: | Computing Science |
| Thesis type: | Master's Internship Report |
| Language: | English |
| Date Deposited: | 23 Jul 2025 07:42 |
| Last Modified: | 03 Sep 2025 08:43 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/36323 |
Actions (login required)
 |
View Item |