Vasudevan, Aneesh (2025) Can High-Fidelity Photorealism Provided by Unreal Engine 5 Bridge the Synthetic-to-Reality Gap for Improved Multi-Rotor Drone Detection? [USQ Project]
![[img]](https://sear.unisq.edu.au/style/images/fileicons/text.png) |
Text (Project – redacted)
VASUDEVAN_A_Dissertation_Redacted.pdf Download (14MB) |
Abstract
Rising challenges to international privacy, safety, security and peace posed by multi-rotor drone operators has accelerated research and development of counter small Unmanned Aircraft Systems (C-sUAS) systems. As part of this, computer vision (CV) approaches using deep learning (DL) have been typically utilised as an effective method for the detection of multi-rotor drones, by way of electro-optical (EO) camera detection systems. Since large datasets are often required to train and validate CV models, synthetically generated image datasets are a promising method that have been explored to meet this demand. Unreal Engine 5 (UE5) is a rendering engine that has grown in prominence due to achievable high-fidelity photorealism not only for gaming experiences, but also digital content creation such as imagery. This is made possible with Lumen, a new lighting system by UE5, which is a global illumination (GI) and reflections system, that simulates accurate lighting, soft shadows and physics-based material properties.
Training drone direction models exclusively with synthetic data, even with high-levels of photorealism, has long caused an issue called synthetic-to-reality gap, due to inherent differences between real-world and synthetic data. Strategies such as domain randomisation are commonly practiced to minimise this issue, however, the question still remains if high-fidelity photorealism in generated training imagery by a state-of-the-art rendering platform such as UE5, in tandem with other strategies can effectively bridge the synthetic-to-reality gap for improved drone detection. Preliminary results of this research have indicated that UE5 can in fact can assist in bridging this gap, however, future research is still required.
|
Statistics for this ePrint Item |
| Item Type: | USQ Project |
|---|---|
| Item Status: | Live Archive |
| Faculty/School / Institute/Centre: | Current – Faculty of Health, Engineering and Sciences - School of Engineering (1 Jan 2022 -) |
| Supervisors: | Lowe, Tobias |
| Qualification: | Bachelor of Engineering (Honours) (Mechatronic) |
| Date Deposited: | 18 Mar 2026 05:46 |
| Last Modified: | 18 Mar 2026 05:46 |
| Uncontrolled Keywords: | Unreal Engine 5; Drone Detection; Synthetic Data; Photorealism; Lighting; Fidelity |
| URI: | https://sear.unisq.edu.au/id/eprint/53173 |
Actions (login required)
 |
Archive Repository Staff Only |
