Applying Differential Privacy for Real-Time Monitoring Systems via Compressive Sensing

Real-time monitoring systems capturing image/video feeds are commonplace in all public spaces, and preserving privacy in such transmission channels is most important. This paper proposes a lightweight differential privacy preserving image compression method for real-time monitoring systems targeted at resource-constrained hardware-based sensors. We use the core principles of compressive sensing augmented with the controlled encryption of its measurement matrix to achieve fine control over the amount of privacy in any real-time monitoring system. Our scheme ensures that privileged authorized parties can decompress and extract the original images while unauthorized entities obtain only a custom amount of privacy-preserved images, as per design. The synthesis results show that the proposed $\textsf {CoS-DiP}$ utilized slightly more area than the standard BCS method. BCS provides only compression, whereas our proposed model provides compression and privacy preservation based encryption. Furthermore, $\textsf {CoS-DiP}$ utilized 91% less area and 97.8% less power compared to standard JPEG compression up to DCT (compression alone). We also presented a comprehensive security and experimental analysis of the proposed method and a working prototype on a camera-mounted Raspberry Pi board, transmitting real-time image feed to a computer over Ethernet.