SceneLock: Reversible Adversarial Learning for Camera-Based Autonomous Driving Protection
Keywords: Data Protection, Autonomous Driving, Camera-Based 3D Perception
Abstract: The advancement of autonomous driving technology hinges on large-scale data collection to train camera-based deep neural network 3D object detectors.
However, these valuable datasets are at risk of unauthorized access and misuse by malicious actors, jeopardizing intellectual property, remote deployment, and the privacy of sensitive information captured during data collection.
We propose a novel reversible adversarial learning framework, referred to as SceneLock, aimed at protecting autonomous driving data from unauthorized use.
Our method conducts adversarial perturbations through a carefully designed Noise Serialization Encoding module (NSE), which significantly degrades image quality and renders the data ineffective for unauthorized artificial intelligence models and manual annotation.
To ensure legitimate access remains unaffected, we integrate advanced image steganography to embed perturbation values within the images.
Furthermore, authorized users can extract these values using appropriate decryption tools through the Noise Serialization Decoding module (NSD) to restore the original high-quality images.
Experimental results demonstrate that our approach effectively safeguards data integrity against unauthorized use while maintaining availability for legitimate purposes.
This dual-layer protection highlights the potential of our method to enhance data security in the autonomous driving domain.
Primary Area: applications to computer vision, audio, language, and other modalities
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: 6250
Loading