Abstract
Drone-based applications involving real-time video analytics are emerging to serve diverse situational awareness use cases ranging from precision agriculture to disaster response. Hence, there is a need to study robust security measures to ensure the integrity of information and safeguard communication, as well as data transmission in drone video analytics. In this paper, we investigate methods to enhance the security management of drone video analytics in terms of reliability and integrity within realistic settings using testbed resources in the NSF-supported AERPAW infrastructure. Specifically, we study security mechanisms to model and detect threats such as Replay, Packet Injection, and Physical Capture attacks caused by situations in dynamic and potentially adversarial network environments. In addition, we generate balanced datasets through Generative Adversarial Networks (GAN) to address challenges posed by unbalanced datasets that are common when applying machine learning models for attack detection impacting drone video analytics traffic. Our experimental environment in AERPAW involves a setup for secure communication through a MAVLink-based (open-standard) drone communication protocol that uses continuous authentication via digital signatures. Our experiment results compare the efficiency gains achieved through secure MAVLink-based communication with unsecured counterparts, examining factors such as packet encryption, digital signatures, and nonces. Further, our results provide valuable insights into the adaptability of security mechanisms for drone video analytics within realistic environments.