Computer Vision Techniques for Military Surveillance Drones

Abstract

Commercial unmanned aerial vehicles (UAVs), also referred to as drones, have proliferated recently, raising concerns about security threats and the need for effective countermeasures. To address these concerns, various technologies have been explored, including radar, acoustics, and RF signal analysis. However, computer vision, particularly deep learning approaches, has emerged as a robust and widely used method for autonomous drone identification. The goal of this research is to create an autonomous drone identification and surveillance system that makes use of a mix of static wide-angle cameras and a lower-angle camera placed on a revolving turret. To optimize memory and processing time, we suggested a novel multi-frame DL identification model. In this approach, the frames captured by the turret's magnified camera are stacked on top of the frames from the wide-angle still camera. Utilizing this technique, we can create an efficient pipeline that conducts initial identification of small-sized aerial invaders on the primary picture plane and identification on the expanded image plane at the same time. This approach significantly reduces the computational burden associated with detection algorithms, making it more resource-efficient. Furthermore, we present the complete system architecture, which includes DL classification frameworks, tracking algorithms, and other essential components. By integrating these elements, we create a comprehensive solution for drone identification and tracking. The system leverages the power of deep learning to accurately classify and track drones in real-time, enabling prompt response and mitigating potential security threats. Overall, this research offers a novel and effective approach to autonomously identify and track drones using computer vision and deep learning techniques. By combining static and dynamic camera perspectives and employing a multi-frame detection method, we provide a resource-efficient solution for drone identification. This work contributes to the ongoing efforts in enhancing security measures against potential drone-related risks