Detecting Spying Activities from the Sky via Deep Learning

The increasing prevalence of unmanned aerial vehicles (UAVs), or drones, has raised serious privacy concerns from the general public due to their pervasive reachability and rich sensing capabilities. In recent years, multiple systems have been proposed to enable drone detection for the general public. However, these systems mainly focus on the detection of nearby drones, but are not able to tell whether the drones are performing unauthorized activities or not. In this paper, we propose a novel solution for the detection of spying activities from drones, which can effectively identify if a drone is spying on a specific victim. Our solution is designed with a deep learning-enabled multi-level analysis by exploring the characteristics of drones' first-personview (FPV) communication channel. Both indoor and outdoor environments are supported in our solution with low-cost off-the-shelf hardware (under $50). A prototype implementation is provided and evaluated under different conditions. Our experimental results demonstrated that our solution is able to achieve high accuracy under different conditions.