Invisibility Stickers Against LiDAR: Adversarial Attacks on Point Cloud Intensity for LiDAR-based Detection
Keywords: LiDAR-based detection, adversarial examples, deep learning
Abstract: Point cloud detection is crucial in applications such as autonomous driving systems and robotics. These systems utilize onboard LiDAR sensors to capture input point clouds, consisting of numerous three-dimensional coordinate points and their corresponding intensity of laser reflection. Recent studies have proposed various adversarial schemes to highlight the vulnerability of point cloud detectors. However, these studies primarily focused on generating or perturbing the coordinate positions of input points and are hard to attack in the physical world, while largely overlooking the significance of their intensity. Through our exploration, we found that perturbing point cloud intensity poses significant security risks for point cloud object detectors. To the best of our knowledge, we are the first to attack on point cloud intensity and we propose an effective adversarial attack scheme, named I-ADV. Our method employs a voxel partition scheme to enhance physical implementation. To boost attack performance, we incorporate a gradient enhancement technique using 3D angle and distance features, along with an extremum-based gradient fusion strategy. Extensive experimental results demonstrate that by altering only point cloud intensity, our approach achieves state-of-the-art performance across detectors with various input representations, attaining attack success rates between 83.9% and 99.1%. Comprehensive ablation studies confirm the effectiveness and generality of the method’s components. Additionally, comparing different attack schemes underscores the advantages of our point cloud intensity attack method in both performance and real-world applicability.
Primary Area: alignment, fairness, safety, privacy, and societal considerations
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4519
Loading