Synthetic Data for Drone vs. Bird Detection

Abstract

The increasing demand for effective drone detection systems has led to the emergence of DL methods as promising solutions for identifying and tracking unauthorized UAVs. However, these detectors face a variety of challenges, particularly in distinguishing drones from visually similar objects such as birds. To address this issue, DL-based drone detectors require substantial training data, which is costly and time-consuming to obtain, especially with clear annotations for both drones and birds. Therefore, this thesis explores the generation of synthetic data using game engine-based simulations in three-dimensional environments for drone vs. bird detection.