Lens Privacy Sealing: Hardware Solution for Privacy-Preserving Action Recognition
A team of researchers has introduced Lens Privacy Sealing (LPS), an economical hardware method that utilizes adjustable laminating film to obscure camera lenses, enhancing privacy during data collection in RGB surveillance systems. Unlike costly engineered optics or algorithms applied after capture, LPS ensures pre-sensor privacy through physically irreversible stochastic multi-layer scattering. They have also launched the P$^3$AR dataset aimed at privacy-preserving action recognition, which includes extensive replay-captured (P$^3$AR-NTU, 114K videos) and real-world collected (P$^3$AR-PKU) subsets annotated with privacy attributes. To address video quality issues arising from LPS, they developed MSPNet, a single-stage framework featuring Inter-Frame Noise Suppress. This research is available on arXiv under ID 2605.19578.
Key facts
- LPS uses adjustable laminating film to physically obscure camera lenses.
- LPS provides pre-sensor privacy protection at minimal cost.
- Privacy is achieved through stochastic multi-layer scattering that is physically irreversible.
- P$^3$AR dataset includes P$^3$AR-NTU (114K videos) and P$^3$AR-PKU subsets.
- MSPNet is a single-stage framework for handling video degradation from LPS.
- The method addresses privacy concerns in RGB camera-based surveillance systems.
- Existing methods rely on post-capture algorithms that fail during data acquisition.
- The paper is available on arXiv with ID 2605.19578.
Entities
Institutions
- arXiv