Infrared Adversarial Patches with Learnable Shapes and Locations in the Physical World

Infrared Adversarial Patches with Learnable Shapes and Locations in the Physical World

연구 분야: Verification

학회: International Journal of Computer Vision

Owing to the extensive application of infrared object detectors in the safety-critical tasks, it is necessary to evaluate their robustness against adversarial examples in the real world. However, current few physical infrared attacks are complicated to implement in practical application because of their complex transformation from the digital world to physical world. To address this issue, in this paper, we propose a physically feasible infrared attack method called “infrared adversarial patches”. Considering the imaging mechanism of infrared cameras by capturing objects’ thermal radiation, infrared adversarial patches conduct attacks by attaching a patch of thermal insulation materials on the target object to manipulate its thermal distribution. To enhance adversarial attacks, we present a novel aggregation regularization to guide the simultaneous learning for the patch’s shape and location on the target object. Thus, a simple gradient-based optimization can be adapted to solve for them. We verify infrared adversarial patches in different object detection tasks with various object detectors. Experimental results show that our method achieves more than 90% Attack Success Rate (ASR) versus the pedestrian detector and vehicle detector in the physical environment, where the objects are captured in different angles, distances, postures, and scenes. More importantly, infrared adversarial patch is easy to implement, and it only needs 0.5 h to be manufactured in the physical world, which verifies its effectiveness and efficiency. Another advantage of our infrared adversarial patches is the ability to extend to attack the visible object detector in the physical world. As a consequence, we can simultaneously perform the infrared and visible physical attacks by a unified adversarial patch, which shows the good generalization.