Athena Seminar Series: Banshee: Target Switch Attacks on Gimbal-Stabilized Visual Tracking Systems via Acoustic Injection

Presenter: Jiarui Li

Gimbal-stabilized visual tracking is widely used in autonomous systems such as UAVs. While prior work shows that acoustic signals can disturb gimbal internals, their impact on end-to-end tracking applications remains unclear. We present Banshee, the first physically realizable attack that induces target switching in UAV visual tracking by exploiting acoustic vulnerabilities in gimbal-camera systems. Banshee generates acoustic signals that cause directionally biased camera motion, breaking inter-frame target associations and steering the tracker toward an attacker-chosen object, leading to target switch or target loss. We achieve 93.6% success in simulation and 95.5% in real-world benchtop and in-flight experiments. Our results reveal a practical cross-domain vulnerability between acoustics and vision, highlighting the need for more robust system designs.