Infrared Adversarial Car Stickers

Infrared physical adversarial examples are of great significance for studyingthe security of infrared AI systems that are widely used in our lives such asautonomous driving. Previous infrared physical attacks mainly focused on 2Dinfrared pedestrian detection which may not fully manifest its destructivenessto AI systems. In this work, we propose a physical attack method againstinfrared detectors based on 3D modeling, which is applied to a real car. Thegoal is to design a set of infrared adversarial stickers to make cars invisibleto infrared detectors at various viewing angles, distances, and scenes. Webuild a 3D infrared car model with real infrared characteristics and propose aninfrared adversarial pattern generation method based on 3D mesh shadow. Wepropose a 3D control points-based mesh smoothing algorithm and use a set ofsmoothness loss functions to enhance the smoothness of adversarial meshes andfacilitate the sticker implementation. Besides, We designed the aluminumstickers and conducted physical experiments on two real Mercedes-Benz A200Lcars. Our adversarial stickers hid the cars from Faster RCNN, an objectdetector, at various viewing angles, distances, and scenes. The attack successrate (ASR) was 91.49and no sticker were only 6.21of the designed stickers against six unseen object detectors such as YOLOv3 andDeformable DETR were between 73.35attack performance across detectors.