Computation-Efficient and Recognition-Friendly 3D Point Cloud Privacy Protection

2025-03-20

AI Summary

3D point clouds in autonomous driving and robotics pose significant geometric-structure privacy risks—an issue largely unaddressed in prior work. Method: This paper formally defines the 3D point cloud privacy problem and proposes the first efficient, recognition-friendly privacy-preserving framework. It introduces (1) PointFlowGMM, a streaming generative model that combines Gaussian mixture latent-space projection with angular similarity loss to achieve geometric obfuscation and model lightweighting; and (2) an orthogonal random rotation transform that preserves inter-class discriminability while protecting individual geometric privacy. Results: Encrypted point clouds achieve classification and segmentation accuracy comparable to original data, reduce model size by 84% (767 MB → 120 MB), significantly lower inference overhead, and enable downstream task execution without access to raw point clouds.

Technology Category

• Machine Learning: Privacy
• Computer Vision: Bias, Fairness & Privacy
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

3D point cloud has been widely used in applications such as self-driving cars, robotics, CAD models, etc. To the best of our knowledge, these applications raised the issue of privacy leakage in 3D point clouds, which has not been studied well. Different from the 2D image privacy, which is related to texture and 2D geometric structure, the 3D point cloud is texture-less and only relevant to 3D geometric structure. In this work, we defined the 3D point cloud privacy problem and proposed an efficient privacy-preserving framework named PointFlowGMM that can support downstream classification and segmentation tasks without seeing the original data. Using a flow-based generative model, the point cloud is projected into a latent Gaussian mixture distributed subspace. We further designed a novel angular similarity loss to obfuscate the original geometric structure and reduce the model size from 767MB to 120MB without a decrease in recognition performance. The projected point cloud in the latent space is orthogonally rotated randomly to further protect the original geometric structure, the class-to-class relationship is preserved after rotation, thus, the protected point cloud can support the recognition task. We evaluated our model on multiple datasets and achieved comparable recognition results on encrypted point clouds compared to the original point clouds.

Problem

• Addresses privacy leakage in 3D point clouds
• Proposes efficient privacy-preserving framework PointFlowGMM
• Maintains recognition performance with reduced model size

Innovation

• Flow-based generative model for 3D point cloud privacy
• Angular similarity loss to obfuscate geometric structure
• Orthogonal rotation preserves class-to-class relationships