Physical 3D Adversarial Attacks against Monocular Depth Estimation in Autonomous Driving
This addresses a critical safety problem for autonomous driving systems by exposing vulnerabilities to physical attacks, representing a novel advancement beyond previous 2D methods.
The paper tackles the vulnerability of monocular depth estimation in autonomous driving by proposing 3D Depth Fool, a 3D texture-based adversarial attack that causes depth errors exceeding 10 meters in real-world tests.
Deep learning-based monocular depth estimation (MDE), extensively applied in autonomous driving, is known to be vulnerable to adversarial attacks. Previous physical attacks against MDE models rely on 2D adversarial patches, so they only affect a small, localized region in the MDE map but fail under various viewpoints. To address these limitations, we propose 3D Depth Fool (3D$^2$Fool), the first 3D texture-based adversarial attack against MDE models. 3D$^2$Fool is specifically optimized to generate 3D adversarial textures agnostic to model types of vehicles and to have improved robustness in bad weather conditions, such as rain and fog. Experimental results validate the superior performance of our 3D$^2$Fool across various scenarios, including vehicles, MDE models, weather conditions, and viewpoints. Real-world experiments with printed 3D textures on physical vehicle models further demonstrate that our 3D$^2$Fool can cause an MDE error of over 10 meters.