Enhancing Counter Drone Operations Through Human-AI Collaboration: A Hierarchical Decision-Making Framework

In recent years, drones have become popular for various applications, including surveillance and delivery. Detecting and classifying drones is crucial for security and regulatory purposes, as well as for improving the efficiency of drone operations. In this work, we propose a human-AI collaboration framework for drone detection, tracking, and classification using multiple sensors, such as ground radar and ground pan-tilt-zoom (PTZ) camera. Our approach combines traditional signal processing and machine learning to classify drones using radar tracks, and a deep learning-based model for detecting and classifying drones from video data. However, due to the limitations of supervised learning, there are always scenarios where these models fail, such as unseen/untrained backgrounds, new aircraft types or signatures. To address this issue, we developed a human-like model and proposed a methodology that involves object detection and a referral policy to trigger a simulated human review of the event. We demonstrate the effectiveness of our human-AI collaboration framework through extensive simulations utilizing data collected from ground radar, and comparing it with standalone AI systems. We also evaluate the effectiveness of various referral policies for different human decision models. In conclusion, the proposed human-AI collaboration framework for drone detection, tracking, and classification has the potential to improve the efficiency and accuracy of drone operations, while ensuring regulatory compliance and enhancing security. Multiple sensors and a referral policy is applied in the proposed framework to trigger human review of events. This is an important step towards addressing the limitations of supervised learning in scenarios where traditional models fail, and towards achieving the full potential of drones in various applications.