Generating Synthetic Data for Deep Learning-based Drone Detection

Tamara R. Lenhard, Andreas Weinmann, Eva Brucherseifer

Abstract: Drone detection is an important yet challenging task in the context of object detection. The development of robust and reliable drone detection systems requires large amounts of labeled data, especially when using deep learning (DL) models. Unfortunately, acquiring real data is expensive, time-consuming, and often limited by external factors. This makes synthetic data a promising approach to addressing data deficiencies. In this paper, we present a data generation pipeline based on Unreal Engine4.25 and Microsoft AirSim, designed to create synthetic labeled data for drone detection using three-dimensional environments. As part of an ablation study, we investigate the potential use of synthetic data in drone detection by analyzing different training strategies, influencing factors, and data generation parameters, specifically related to the visual appearance of a drone.