Single-Camera-Based 3D Drone Trajectory Reconstruction for Surveillance Systems

Abstract

Drones have been utilized in various fields, but the number of drones being used illegally and for illegal and hazardous purposes has recently increased. These misuses have given rise to the development of an anti-drone system, C-UAS, comprising two key steps: engagement and detection. This detection step plays a crucial role in reconstructing the drone’s trajectory for the subsequent engagement step. Therefore, we focus on a trajectory reconstruction approach that utilizes external data, such as CCTV view images, instead of internal drone data. While numerous methods have been explored for estimating 3D drone trajectory using multiple sensors, they are often unsuitable for surveillance systems. In this study, we use a calibrated single camera suitable for surveillance systems, leveraging the relationship between 2D and 3D spaces. We use a drone tracker to automatically track 2D images. The tracked images are used to estimate the 2D rotation of the drone in the image through principal component analysis (PCA). By combining the estimated 2D drone positions with the actual length, we geometrically infer the 3D drone trajectories. We also develop synthetic 2D and 3D drone datasets to address the lack of public drone datasets. Additionally, a real-world 3D dataset is generated and made available for public use. The experimental results demonstrate that the proposed method can accurately reconstruct drone trajectories in 3D space. With a MAE of 5.66 and RMSE of 7.93 showing low error rates, these findings validate the practical value and potential of our framework as a single-camera-based surveillance system.