Guarding the Senses: Unveiling Cybersecurity Solutions through the Exploration of Image and Sound Sensor Design Limitations

The proliferation of Internet of Things (IoT) devices in modern society has seen an increasing dependence on sensors, particularly cameras and microphones, to enhance functionality and user experience. However, this advancement is not without its pitfalls. This dissertation explores the potential security vulnerabilities that these sensors introduce. Previous work has shown that by targeting microphone sensors, they can subvert voice-controllable systems. Similarly, by manipulating camera sensors, attackers can compromise the safety of autonomous vehicles. Through rigorous research, we have discovered novel attacks tailored to these sensors: a remote and inaudible assault on microphones and a subtle, unnoticeable compromise of cameras. Recognizing the implications of these vulnerabilities, we have also designed and proposed effective defenses against these sophisticated attacks, aiming to bolster the security of IoT devices in our interconnected world. Our research can be divided into three parts: (i) Inaudible attack against voice-controllable system (VCS) via compromising microphone. This part includes three of our works: one published work named Near-Ultrasound Inaudible Trojan (Nuit): Exploiting Your Speaker to Attack Your Microphone, and one submitted work Surface Obstacles as Speakers (SOS): Waging Inaudible Attacks against Voice Control Systems Behind Surface Obstacles; (ii) Invisible attack against Autonomous driving AI model via compromising camera. In this part includes our submitted work Moiré Injection Attack (MIA): Compromising Autonomous Vehicle Safety via Exploiting Camera's Color Filter Array (CFA) to Inject Hidden Traffic Sign. (iii) In this part we also include Alpha Channel Attack (ACA), Ignoring is Attacking: The Universal Targeted No-box Attack on Computer Vision Models through Alpha Channel Oversight.