Seoul — The ACM WiSec Conference — With the rise of drones posing a threat to critical infrastructure, public safety, and privacy due to their affordability and convenient operation, a security expert introduced a new drone detection system on Tuesday that integrates both video and audio detection solutions, enhancing surveillance capabilities against unauthorized drone activity.
Ildi Alla, a researcher at Inria Lille-Nord Europe, a research center affiliated with the University of Lille, presented his team’s work to the international audience at the 17th ACM Conference on Security and Privacy in Wireless and Mobile Networks (ACM WiSec).
The researcher, drawing from a research paper titled ‘From Sound to Sight: Audio-Visual Fusion and Deep Learning for Drone Detection,’ emphasized the significance of precisely identifying drones. With their increasing misuse posing a growing threat to infrastructure and public security, accurate detection methods are crucial.
For example, according to reports from foreign news outlets, two drones collided with separate oil refineries in Russia last year. One of these incidents resulted in a fire, which was subsequently extinguished. While the damage was not severe, the proximity of the oil refineries to one of the country’s largest oil export gateways, the port of Novorossiysk, raised concerns and alarm.
The security expert highlighted the limitations of the current single-sensor system for detecting drone activity—specifically, that it relies on a single detector and, due to this, often fails to achieve the accuracy necessary to be optimally effective. While an infrared camera, also known as a thermal imaging camera, can be effective in visually challenging conditions such as darkness, its effectiveness diminishes over longer distances. On the other hand, audio-based methods can capture distinctive signatures and perform well in low-visibility environments. However, they may encounter challenges in noisy conditions, making it difficult for users to detect drones effectively.
To address the challenges inherent in various detection methods, the study concentrated on developing an anti-drone system that harnesses the strengths of both infrared imaging and audio sensors. The security expert explained that the team conducted a comparative analysis of deep learning models for audio and video recognition. They then integrated the outputs of their analysis to develop a multi-model detection technique. This approach involved utilizing a dataset consisting of 375 audio files and 20,688 annotated images containing drones. The expert revealed that their detection framework achieved an impressive accuracy rate of 96.02%, outperforming existing drone detection methods.
The expert detailed plans to integrate radio frequency (RF) sensors into their system for improved detection in their future work. RF-based detection involves capturing radio-frequency signals exchanged between the drone and its controller, which can help identify unique signals. However, implementing RF detection in scenarios involving multiple drones could pose challenges. The researcher explained, “We are currently collecting data and incorporating radio frequency for improvement. We plan to further test it with a variety of techniques.”
