SynDroneVision: A Synthetic Dataset for Image-Based Drone Detection

TL;DR

SynDroneVision introduces a comprehensive synthetic RGB drone-detection dataset generated with Colosseum and Unreal Engine 5 to address data scarcity in surveillance contexts. The work demonstrates that training with a hybrid mix of synthetic and real data improves YOLO-based detection performance and robustness, while maintaining competitive results when trained on synthetic data alone. Key contributions include a detailed data-generation pipeline, diverse environment and illumination variations, post-processing augmentations, and an extensive evaluation across multiple YOLO architectures and out-of-distribution datasets. The practical impact lies in lowering data-collection costs and enabling rapid, scalable training of robust drone detectors for real-world security applications, with public release planned upon acceptance.

Abstract

Developing robust drone detection systems is often constrained by the limited availability of large-scale annotated training data and the high costs associated with real-world data collection. However, leveraging synthetic data generated via game engine-based simulations provides a promising and cost-effective solution to overcome this issue. Therefore, we present SynDroneVision, a synthetic dataset specifically designed for RGB-based drone detection in surveillance applications. Featuring diverse backgrounds, lighting conditions, and drone models, SynDroneVision offers a comprehensive training foundation for deep learning algorithms. To evaluate the dataset's effectiveness, we perform a comparative analysis across a selection of recent YOLO detection models. Our findings demonstrate that SynDroneVision is a valuable resource for real-world data enrichment, achieving notable enhancements in model performance and robustness, while significantly reducing the time and costs of real-world data acquisition. SynDroneVision will be publicly released upon paper acceptance.

Figures (8)

• Figure 1: A selection of synthetic images captured from diverse virtual environments -- University Site (row 1), Venetian City (row 2), Rural Australia (row 3), Factory Grounds (row 4), and Modular Cityscape (last row) -- demonstrating SynDroneVision's diversity in terms of environmental conditions and camera perspectives (top-down, ground-level, and bottom-up).
• Figure 2: Drone models from Unreal:QuadcopterPackUnreal:MilitaryDronesUnreal:DronePack employed in the generation of the SynDroneVision dataset.
• Figure 3: Position distribution of drones within the SynDroneVision dataset. Regions of high frequency are shown in yellow, while areas with no data points are indicated in blue.
• Figure I: Folder configuration of the SynDroneVision dataset.
• Figure II: Customized camera perspectives and lighting configurations tailored to each environment. The camera fields of view (FOVs) correspond to the following environments (arranged from left to right, top to bottom): University Site (rows 1-3), Venetian City (rows 4-5), Farming Grounds (row 6), Rural Australia (row 7), City Park (images 1-5, row 8), Factory Grounds (image 6, row 8; images 1-6, row 9), Urban Downtown (images 1-6, row 10; images 1-4, row 11), and Modular Cityscape (images 5-6, row 11; images 1-6, last row).