REAL-TIME DRONE DETECTION USING CONVOLUTIONAL NEURAL NETWORKS IN COMPLEX ENVIRONMENTS

Abstract

With the relatively quick growth of the accessibility and affordability of Unmanned Aerial Vehicles (UAVs) and drones, opportunities for UAVs exist within many sectors, including defence, agriculture, surveillance, and logistics. Additionally, while these UAVs can be used for a variety of purposes, the mass deployment of these vehicles has raised serious concerns regarding safety, security, and privacy. As an example, in low-altitude airspace, effective drone detection has become more challenging due to the drone's small size, high speed, and ability to closely resemble other objects such as birds in a typical urban environment. Additionally, through the introduction of new environmental conditions like background clutter, varying levels of illumination, and difficult weather conditions, the effective use of traditional drone detection technology (i.e., radar, acoustic, machine learning with manually created feature sets) is increasingly susceptible to being unable to operate effectively in real-world scenarios. This research proposes a novel drone detection system that utilizes a Convolutional Neural Network (CNN) to detect small drones in low-altitude airspace. The CNN detection algorithm will learn hierarchical features directly from the raw images rather than manually creating the features. When benchmarked against traditional technology, the CNN system demonstrates improved levels of accuracy and robustness. The results of this work will support the development of cost-effective, easily scalable, and rapid responding UAV detection technologies to improve the safety of the public and increase national security, as well as protect critical infrastructure.